Mechanisms and Machine Science

Eusebio E. Hernandez
Sajjad Keshtkar
S. Ivvan Valdez Editors

Industrial
and Robotic
Systems
LASIRS 2019
Mechanisms and Machine Science

Volume 86

Series Editor
Marco Ceccarelli, Department of Industrial Engineering, University of Rome Tor
Vergata, Roma, Italy

Editorial Board
Alfonso Hernandez, Mechanical Engineering, University of the Basque Country,
Bilbao, Vizcaya, Spain
Tian Huang, Department of Mechatronical Engineering, Tianjin University,
Tianjin, China
Yukio Takeda, Mechanical Engineering, Tokyo Institute of Technology, Tokyo,
Japan
Burkhard Corves, Institute of Mechanism Theory, Machine Dynamics and
Robotics, RWTH Aachen University, Aachen, Nordrhein-Westfalen, Germany
Sunil Agrawal, Department of Mechanical Engineering, Columbia University
New York, NY, USA
This book series establishes a well-defined forum for monographs, edited Books,
and proceedings on mechanical engineering with particular emphasis on MMS
(Mechanism and Machine Science). The final goal is the publication of research that
shows the development of mechanical engineering and particularly MMS in all
technical aspects, even in very recent assessments. Published works share an
approach by which technical details and formulation are discussed, and discuss
modern formalisms with the aim to circulate research and technical achievements
for use in professional, research, academic, and teaching activities.
This technical approach is an essential characteristic of the series. By discussing
technical details and formulations in terms of modern formalisms, the possibility is
created not only to show technical developments but also to explain achievements
for technical teaching and research activity today and for the future.
The book series is intended to collect technical views on developments of the
broad field of MMS in a unique frame that can be seen in its totality as an
Encyclopaedia of MMS but with the additional purpose of archiving and teaching
MMS achievements. Therefore, the book series will be of use not only for
researchers and teachers in Mechanical Engineering but also for professionals and
students for their formation and future work.
The series is promoted under the auspices of International Federation for the
Promotion of Mechanism and Machine Science (IFToMM).
Prospective authors and editors can contact Mr. Pierpaolo Riva (publishing
editor, Springer) at: pierpaolo.riva@springer.com
Indexed by SCOPUS and Google Scholar.

More information about this series at http://www.springer.com/series/8779

Eusebio E. Hernandez •

Sajjad Keshtkar S. Ivvan Valdez

•

Editors

Industrial and Robotic

Systems
LASIRS 2019

123
Editors
Eusebio E. Hernandez Sajjad Keshtkar
ESIME Ticoman School of Engineering and Science
Instituto Politecnico Nacional Tecnologico de Monterrey
Mexico City, Mexico Altamira, Mexico

S. Ivvan Valdez
CONACYT-CENTROGEO
Mexico City, Mexico

ISSN 2211-0984 ISSN 2211-0992 (electronic)

Mechanisms and Machine Science
ISBN 978-3-030-45401-2 ISBN 978-3-030-45402-9 (eBook)
https://doi.org/10.1007/978-3-030-45402-9
© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature
Switzerland AG 2020
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt from
the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein or
for any errors or omissions that may have been made. The publisher remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

This book compiles selected and reviewed papers of the Latin American
Symposium on Industrial and Robotic Systems 2019 (LASIRS 2019), from October
30 to November 01, 2019, Tecnologico de Monterrey, Campus Tampico, Mexico.
This conference is a second calling for the community interested in robotics, in
particular, for the Latin American community interested in robotics engineering
applications from algorithms, simulations, and design, to prototypes and imple-
mentations. A first meeting took place at Guanajuato, México, in 2017, under the
name of School of Optimization, Simulation and Numerical Methods in Robotics,
and as the name suggests, this first experience aimed to bring together researchers
and students with a computer science profile but devoted to solving robotics
problems; nevertheless, the experience was so enriching that make us noticed that
the frontiers in robotics problems are not strict, nor in the nature of the problems
either in the necessary perspectives to solve them. Hence, for this second edition,
we convoke researchers in a more broad spectrum of robotics, keeping in mind
several goals: to focus on the applicability of the methods and proposals presented
in the Symposium, the aim to present different perspectives and complementary
disciplines around robotics, the aim to include introduction and tutorial talks for
undergraduate students as an essential part of the symposium in order to promoting
this beautiful area of knowledge to the new generations, and the last but not least,
and possibly our main challenge for future editions, to promote the meeting among
the large community, we know, exists in Latin America. In these last two points, we
are aware of the talented community we have; for instance, very often Latin
American teams are participants and winners of international robotics contest, so
our future vision of this meeting is an academic festivity around robotics with Latin
American students as the special guest.
In this compilation, the reader can find several papers, with topics from new
algorithms, design, simulation, and implementation of industrial and robotics sys-
tems; many of them are valuable by their tutorial content even if they introduce new
material, other are new rigorous research papers, and some others are descriptions
of novel robotics implementation. We expect that the reader finds this compilation

v
vi Preface

useful, that is, at the same time, an invitation to enjoy the broad and fascinating
industrial–robotics world.
We would like to take this opportunity to extend our recognition to the scientific
and organizing committee members, session chairs, reviewers, and conference
participants for the scientific success of this event. Their active participation in the
discussions and the numerous interesting presentations were vital to achieving the
scientific meeting objectives. We are grateful to Tecnologico de Monterrey,
Campus Tampico, for hosting the event. We would also like to thank the auspices
and support of The International Federation for the Promotion of Mechanisms and
Machine Science (IFToMM) and the financial support of the National Council of
Science and Technology (CONACyT), Mexico, through grant 299374.

Eusebio E. Hernandez
Sajjad Keshtkar
S. Ivvan Valdez
Organization

LASIRS 2019 Organizing Committee

Manuel Martínez Martínez School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
Irandi Gutierrez Carmona School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
Sajjad Keshtkar School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
Juan Gabino Díaz Martínez School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
Roberto Abraham Valdivia School of Engineering and Science, Tecnologico
Beutelspacher de Monterrey, Altamira, Mexico
Leopoldo Morales Gómez School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
León Felipe Guevara Chávez School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
Jorge Álvarez Ramírez School of Engineering and Science, Tecnologico
de Monterrey, Altamira, Mexico
Miguel Ángel Contreras School of Engineering and Science, Tecnologico
Enríquez de Monterrey, Altamira, Mexico

LASIRS 2019 Scientific Committee

Alexander Poznyak CINVESTAV of National Polytechnic Institute,
Mexico
Eusebio Hernández National Polytechnic Institute, Mexico
Rafiq Ahmad University of Alberta, Canada
Hirohisa Kojima Tokyo Metropolitan University, Japan
Mario Acevedo Gómez Panamerican University, Mexico
Juan Carlos Jáuregui Correa Autonomous University of Querétaro, Mexico
Héctor Becerra Fermín Mathematics Research Center, Mexico

vii
viii Organization

Edgar Alfredo Portilla Flores National Polytechnic Institute, Mexico

Sajjad Keshtkar Tecnologico de Monterrey, Mexico
Ivvan Valdez Peña CentroMet-CONACyT, Mexico
Adriana Vargas Martínez Tecnologico de Monterrey, Mexico
Jorge Lozoya Tecnologico de Monterrey, Mexico
Armando Román Tecnologico de Monterrey, Mexico
Irandi Gutierrez Carmona Tecnologico de Monterrey, Mexico
Contents

Nanotechnology Applications in Industry and Heat Transfer . . . . . . . . . 1

Sina Razvarz, León Felipe Guevara Chavez, and Cristóbal Vargas-Jarillo
Applications of Fuzzy Logic, Artificial Neural Network
and Neuro-Fuzzy in Industrial Engineering . . . . . . . . . . . . . . . . . . . . . . 9
Raheleh Jafari, Miguel A. Contreras, Wen Yu, and Alexander Gegov
Foundation of Z-Numbers and Engineering Applications . . . . . . . . . . . . 15
Sina Razvarz, Felipe Hernández-Rodríguez, Raheleh Jafari,
and Alexander Gegov
Force Balancing of the 2RRR Planar Parallel Manipulator
via Center of Mass Acceleration Control Using Fully
Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
Mario Acevedo and María T. Orvañanos-Guerrero
Approximation of the Workspace of a Cable-Driven Parallel
Robot with a Movable Gripper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Larisa Rybak, Elena Gaponenko, and Dmitry Malyshev
Design of a New Game for Teaching Assembly Process . . . . . . . . . . . . . 44
Tianxiong Wang and Rafiq Ahmad
An Improved Robot Path Planning Algorithm for a Novel
Self-adapting Intelligent Machine Tending Robotic System . . . . . . . . . . 53
Feiyu Jia, Juan Tzintzun, and Rafiq Ahmad
Stability Analysis of a Rotor Systems with Flow Forces . . . . . . . . . . . . . 65
Jaime Cruz Cruz, Miguel Toledo Velázquez, and Oliver M. Huerta Chávez
Flutter Analysis of Rotor Based on a Fluid–Structure Method . . . . . . . 76
Jaime Cruz Cruz, Miguel Toledo Velázquez, Oliver M. Huerta Chávez,
Gibran Jalil Garnica Castro, and Rafael Sánchez López

ix
x Contents

Amphibious Propeller Optimization for Aerial

and Underwater Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
C. Genaro Carreon, C. Oliver M. Huerta, M. Alfredo Arias,
C. Sergio Torres, and José Ángel L. Ortega H.
Trajectory Tracking of Multiple Quadrotors While Maintaining
Time-Varying Spatial Formations via Synchronization . . . . . . . . . . . . . 97
C.-A. Arteaga-Escamilla, R. Castro-Linares, and J. Álvarez-Gallegos
Modeling, Simulation and Analysis of the Characteristics
of the Non-conventional Vertical Axis Wind Turbine with Variable
Pitch Angle that Allows to Establish Different Control Strategies . . . . . 109
Jesús García-Barrera, Jorge Díaz-Salgado, Oliver M. Huerta-Chávez,
and Rafael Sánchez-López
Numerical Simulation of the Formation, Evolution and Control
of a Methanol Spray Generated by a Pressure Swirl Atomizer
During the Combustion Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
G. Guevara-Morales, O. M. Huerta-Chavez, A. Arias-Montaño,
M. Zirion Flores, and I. Castorena Solache
Control of a Stewart-Gough Platform for Earthquake Ground
Motion Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
Erick Alvarado Requena, Antonio Estrada, Gengis Toledo Ramírez,
Noe Reyes Elias, Jorge Uribe, and Berenice Rodríguez
Flow Control on a Wing Section Through the Implementation
of the SDBD Plasma Actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
R. A. Bernal-Orozco, A. Arias-Montano, and O. Huerta-Chávez
Experimental Platform for Modeling and Control
of Articulated Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
Ana Hernandez-Maldonado, Juan G. Diaz-Martinez, Hirohisa Kojima,
Irandi Gutierrez-Carmona, and Sajjad Keshtkar
Design and Control of a Passive Solar Tracking System Using
a Sky Imager . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
M. Angulo, A. Díaz-Ponce, L. Valentín, R. Valdivia, and S. Keshtkar
Attenuation of the Amplitude Vibration Level by Using Nonlinear
Active Control Techniques Applied to a Hybrid Short Journal
Bearing Using a Particular Rotor-Bearing System Configuration . . . . . 179
D.-E. Rivera-Arreola, J.-C. Gómez-Mancilla, T. Hernández-Cortés,
and H. F. Abundis-Fong
Transient Vibrations Suppression in Parametrically
Excited Resonators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
Miguel Ramírez-Barrios, Fadi Dohnal, and Joaquín Collado
Contents xi

Robust State Estimation for Underactuated Systems Using

Sliding Modes and Attractive Ellipsoid Method . . . . . . . . . . . . . . . . . . . 206
Bonifacio Sanchez, Patricio Ordaz, Ernesto Flores,
and Alexander Poznyak
Modelling and Control of 5 DOF Serial Robot for Object
Capturing and Manipulation Using a Semi-analytical Approach . . . . . . 218
Bolaños Adrian, Sánchez Kevin, Gabino Juan, Gutiérrez-Carmona Irandi,
and Martínez Manuel
Development of a Vision System for a Mobile Manipulator
to Solid Residue Collection in the Food Sector . . . . . . . . . . . . . . . . . . . . 228
Jhoset Méndez Amador, P. A. Niño-Suárez, and Raúl Rivera-Blas
Robust Control of a Master-Slave Manipulator Under
Restricted Task-Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
D. Cruz-Ortiz, I. Chairez, and A. Poznyak
On the Dynamics and Control of a Single-Wheel Robot
with Inertial Locomotion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Mario E. Herrera-Cordero, Manuel Arias-Montiel, Marco Ceccarelli,
and Esther Lugo-González
Analysis and Simulation of the Oscillations in a Pulley-Ribbed
Belt System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
Ernesto Alonso Villegas-Jiménez, Jorge Díaz-Salgado,
Oliver Marcel Huerta-Chávez, Fernando Macedo-Chagolla,
and Edgar Arturo Dávila-Martínez
Attitude Control of Torpedo Anchor with Vectorial Thrust
for Use in a Pseudo-Submersible Platform . . . . . . . . . . . . . . . . . . . . . . . 269
G. J. Garnica-Castro, O. M. Huerta-Chávez, J. Cruz-Cruz,
and M. Toledo-Velázquez
Geometric Parameters and Workspace Optimization
of Sitting-Type Lower Limb Rehabilitation Robot . . . . . . . . . . . . . . . . . 279
Askhat Diveev, Grigory Dubrovin, Dmitry Malyshev,
and Anna Nozdracheva
Comparison of Parallel Versions of SA and GA for Optimizing
the Performance of a Robotic Manipulator . . . . . . . . . . . . . . . . . . . . . . 290
Adán H. Baena, Sergio I. Valdez, Felipe de Jesús Trujillo Romero,
and Moisés M. Montes
Dynamic Modeling and Control of a Pivoting Nozzle . . . . . . . . . . . . . . . 304
Christopher Virgen-Vázquez, Eusebio Eduardo Hernández,
and Armando Oropeza Osornio
xii Contents

A Closed Loop Controller Design Which Can Minimize

the Wind Effects on a Parabolic Solar Tracker . . . . . . . . . . . . . . . . . . . 310
J. Díaz-Salgado, O. M. Huerta-Chávez, E. A. Villegas Jimenez,
E. Enciso-Contreras, L. A. Cortés-Marín, and C. Ortiz-Cil

Author Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327

 Nanotechnology Applications in Industry
and Heat Transfer

Sina Razvarz1(B) , León Felipe Guevara Chavez2 , and Cristóbal Vargas-Jarillo1

1 Departamento de Control Automático CINVESTAV, IPN (National Polytechnic Institute),
07360 Mexico City, Mexico
srazvarz@yahoo.com, cvargas@ctrl.cinvestav.mx
2 School of Engineering and Science, Tecnológico de Monterrey, Altamira,
89600 Tampico, Mexico
Leon.guevara@tec.mx

Abstract. Nanotechnology is stated as the utility of nanomaterials for human

profit. Nanomaterials contain distinct attributes because of their physical as well
as chemical features at the nanoscale. Nanotechnology has a vast domain of usages
and has been utilized in providing new products in the industrial section, electronic
as well as energy. In this paper, the properties of fluids with nanoparticles are
reviewed. Furthermore, the developed models with nanoparticle consortium are
studied.

Keywords: Nanotechnology · Nanomaterials · Industry

1 Introduction
Industrial engineering is one of the fields that artificial neural network, fuzzy logic, and
nano technology have found an extensive implementation area [1–19]. Nanoscale refers
size dimensions between approximately 1 and 100 nm (or more appropriately, 0.2 and
100 nm) because at this scale the properties of materials differ with respect to their
physical properties from a larger scale.
The size dimensions of Nanoscale are among nearly 1 and 100 nm (or more suitably,
0.2 and 100 nm), since at this scale the attributes of materials change concerning their
physical attributes from a greater scale. Every material which contains one or more
dimensions in the nanoscale is stated as nanomaterial [20]. Nanoparticles contain distinct
biological attributes such as tiny size, a great surface-area-to-volume ratio that permits
them to paste, attract as well as to transfer composites like tiny-molecule drugs, proteins
as well as probes having great effectiveness [21].
The field of nanotechnology is built itself as an essential technology for a vast
domain of implementations, so being a first-class preference for science as well as tech-
nology policy growth. Nanotechnology is the implementation of scientific knowledge
for manipulating and control issues at the nanoscale. Nevertheless, there is no defi-
nite answer regarding the usage of nanoparticles and nanomaterials, mostly due to the
shortage of information regarding their attributes and applicability.
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 1–8, 2020.
https://doi.org/10.1007/978-3-030-45402-9_1
2 S. Razvarz et al.

The attributes of materials are not the same at the nanoscale due to two major reasons.
The first reason is that the superficial region of nanomaterials is great and the second
reason is that the quantum effects dominate the treatment of the nanoscale. The physical
nature of the nanoparticles with size less than 20 nm changes more extremely when is
compared with the conventional size materials [22]. The basic attribute for characterizing
nanomaterials are [23]: size propagation, surface region and porosity, and techniques
of synthesis. It is on the basis of their inherent attributes or in the ones acquired using
connecting them with other materials as if possible usages are generated for achievement,
for instance [24]: reduction; weight diminution, and amended functionalities concerning
to the materials.

2 Nanofluids Preparation and Characterization

2.1 Particle Size Distribution

The nanomaterial is a material having an exterior dimension in the nanoscale or with

interior construction at the nanoscale [25] that can reveal new properties in comparison
with the equal material without nanoscale characteristics. The attributes of the materials
may not be the same at the nanoscale due to two major reasons. The first reason is that
the superficial region of nanomaterials is great and the second reason is that the quantum
effects dominate the treatment of the nanoscale.
Applying a standard from 2007, the nanoparticle is stated as a particle having a nom-
inal diameter of less than nearly 100 nm [22]. On a standard from 2008, the nanoparticle
is stated as a particle as if it contains all its three dimensions approximately 100 nm or
minor, also can be mentioned as nano-object [26].
The size dimensions of Nanoscale are among nearly 1 and 100 nm (or more suitably,
0.2 and 100 nm), since at this scale the attributes of materials change concerning their
physical attributes from a greater scale. Every material which contains one or more
dimensions in the nanoscale is stated as nanomaterial. Nanoparticles contain distinct
biological attributes such as tiny size, a great surface-area-to-volume.
The physical attributes of nanoparticles having sizes less than 20 nm can vary more
greatly when are compared with the conventional size materials.
The other perception which is utilized is nanostructured nanoparticles, composed of
particles with structural characteristics less than 100 nm that can impact their physical,
chemical as well as biological characteristics.

2.2 Kind of Nanomaterial

Nanomaterials can cover a heterogeneous domain of materials using classification by

species not entirely clear of controversy.
The European committee demonstrates the below-mentioned classification, on the
basis of different Sustainable and Responsible Investment (SRI) deliberated records [27]:

– Inorganic non-metallic nanomaterials: titanium dioxide (TiO2), silicon dioxide

(SiO2), zinc oxide (ZnO), strontium titanate (SrTiO3), strontium carbonate (SrCO3),
 Nanotechnology Applications in Industry and Heat Transfer 3

indium tin oxide (ITO), antimony tin oxide (ATO), disbismuth trioxide (Bi2O3),
nickel monoxide (NiO) disilver oxide (Ag2O), aluminium nitride (AlN), silicon nitride
(Si3O4), titanium nitride (TiN), titanium carbonitride (TiCN), tungsten carbide (WC),
tungsten sulphide (WS2), etc.;
– Metals and Metal alloys: gold (Au), silver (Ag), platinum (Pt) and palladium (Pd)
alloy, copper (Cu) nanopowders, iron (Fe) nanoparticles, nickel (Ni), cobalt (Co),
aluminium (Al), zinc (Zn), manganese (Mn), molybdenum (Mo), tungsten (W),
lanthanum (La), lithium (Li), rhodium (Rh), etc.
– Carbon-based nanomaterials: fullerenes (C60), carbon nanotubes (CNT, SWCNT,
MWCNT), carbon nanofibers (CNF), carbon black and graphene flaks;
– Nanopolymers and dendrimers: polymer nanoparticles, polymer nanotubes,
nanowires and nanorods, polyglycidylmethacrylate (PGMA) fibres, nanocellulose,
nanostructured polymer-films, polyacrylonitrile, nanostructures (PAN), dendrimers;
– Quantum dots; are inorganic fluorescent semiconductor nanoparticles made of 10–50
atoms having a diameter domain of 2–10 nm [28].
– Nanoshells are nanoparticles made up of a gold Shell encircling a semiconductor
[29].
– Liposomes are the easiest types of nanovectors that are composed of lipids enclosing
the water core.

Nevertheless, there also exist novel kinds of nanomaterials in progress that are mostly
mentioned as the second generation, third generation nanomaterials [27].

2.3 Volume Concentration of Nanofluid

Nanofluids technology can be considered as a novel challenge for the heat transfer fluid
as the thermal conductivity of nanofluids unusually increases at an extremely less volume
concentration [30].
The quantity of nanoparticles needed to prepare nanofluids is computed utilizing the
law of mixture formula. The volume concentration for nanofluids is defined as below
[31]
Vnanopar ticle
ρnanopar ticle
=
Vnanopar ticle  (1)
V
ρnanopar ticle + base f luid ρbase f luid

such that Vnanopar ticle is considered as the volume of nano particle, ρnanopar ticle is
considered as the density of nano particle, Vbase f luid is taken to be the volume of base
fluid, and ρbase f luid is taken to be the density of base fluid.

2.4 Nano Fluid

A nanofluid is a fluid containing nanometer-sized particles, called nanoparticles. These

fluids are engineered colloidal suspensions of nanoparticles in a base fluid. The nanopar-
ticles used in nanofluids are typically made of metals, oxides, carbides, or carbon
nanotubes.
4 S. Razvarz et al.

Nanofluids have been constructed by utilizing two techniques: one step technique and
two step technique. The principal aim of these techniques is to provide a homogeneous
as well as stable solid-liquid mixture, also avoiding the agglomeration, feasible erosion
and clogging etc. In one step technique, the nanoparticles have been constructed and
dispersed into the foundation fluid at the same time. Whereas in a two-step technique,
the nanoparticle has been generated in the first step and dispersed into host fluids in the
second step. Investigation reveals that nanofluids with oxide nanoparticles and carbon
nanotubes can be produced utilizing two-step technique [32]. Two-step technique is
not appropriate for metallic nanoparticles. Under particular circumstances, nanofluids
represent excellent thermo physical features that result in better effect of the thermal
usage.

2.5 Nanofluids Viscosity Characterization

The density of nanofluids increases with an increment in concentration also reduces with
an increment in temperature. The width of nanofluids for every volume concentration is
calculated using Hygrometer. The acquired density information is in contradiction form
also, the qualities are obtained using the thickness connection status developed by Pak
and Choi in 1998 [33–35] for nanofluids.
The density data for water-propylene glycol mix like a function of temperature,
which is extracted from ASHRAE (2009) Hand Book [36].The density of nanofluids is
defined using a blending rule as below
ρnano f luid = (1 − )ρbase f luid + ∅ρbase f luid (2)
Based on the outcomes demonstrated in Fig. 1, the density of nanofluids can forecast
experimental data of density precisely.

Fig. 1. Relation among temperature and density.

2.6 Nanofluid Particular Heat Rate Measurements

The nanofluid particular heat rate can be considered as one of the main characteristics
and also has a major role in impacting heat transfer rate of nanofluids.
 Nanotechnology Applications in Industry and Heat Transfer 5

Particular heat is the rate of heat needed for increasing the temperature of one gram
of nanofluids by one degree centigrade. The heat transfer rate is stated as below

Q = mc p T (3)

such that Q is taken to be the heat transfer rate, m is considered as the mass flow rate,
also T is taken to be the temperature variation of the cooling liquid.
The nanofluid particular heat changes with the nanoparticle concentration in the base
fluid based on Eq. (3) which is generated by Buongiorno in [37].
   
  (1 − ϕ) ρc p f luid + ϕ ρc p nano f luid
c p nano f luid = (4)
(1 − ϕ)(ρ) f luid + ϕ(ρ)nano f luid

such that c p is taken to be the heat capacity of the nanofluid or base fluid.

3 Applications of Nanotechnology in Industry

3.1 Biomedical Applications

Nanoparticles have been utilized for disease recognition, drug transfer, gene therapy of
cancer, and also to prevent other infections.
There exist various nanoparticle-based delivery systems accessible for distinguishing
utilizations like Liposomes (Controlled and targeted drug delivery, targeted gene deliv-
ery); Dendrimers (Targeted drug delivery); Nanoshells (Tumor targeted); Nanowires,
(fullerenes-based derivatives As targeting and imaging agents); Quantum dots (target-
ing and imaging agents) Gold nanoparticles (Targeted delivery and imaging agents);
Carbon nanotube (Drug gene and DNA delivery, tumor targeting).

3.2 Nanotechnology in Agricultural and Food Production

The diversity of nanomaterials, mainly metal-based nanomaterials as well as carbon-

based nanomaterials, are used because of their displacement, agglomeration as well as
affect on formation and evolution of crop plants [38].

3.3 Nanotechnology in the Food Industry

Nanotechnology usages in foods proposed the technologies, which generate added

values, which are greatest required, also contain the most effect on consumers.
Four main kinds of usages of nanoscience in the food industry, are agriculture, food
dealing food wrapping as well as supplements.
Silver nanoparticles have been considered as an efficient concept opposed to
vancomycin-resistant enterococcus as well as extended Extended-spectrum beta-
lactamases [39].
6 S. Razvarz et al.

4 Conclusion
Nanoscience has been broadly utilized in different areas like agricultural industry
because of the specific properties of nanomaterials. In this paper, the properties of fluids
with nanoparticles are reviewed. Furthermore, the developed models with nanoparticle
consortium are studied. The attributes of the nanofluid can be estimated precisely apply-
ing the mixing rule. Viscosity are estimated wisely applying the Einstein model at low
concentrations.

References
1. Razvarz S, Jafari R, Yu, W, Khalili A (2017) PSO and NN modeling for photocatalytic removal
of pollution in wastewater, vol 20, pp 1–6
2. Jafari R, Yu W (2015) Artificial neural network approach for solving strongly degener-
ate parabolic and burgers-fisher equations. In: 12th international conference on electrical
engineering, computing science and automatic control
3. Jafari R, Razvarz S, Gegov A (2018) A new computational method for solving fully fuzzy
nonlinear systems. In: Computational collective intelligence, ICCCI 2018. LNCS, vol 11055.
Springer, Cham, pp 503–512
4. Razvarz S, Jafari R (2017) ICA and ANN modeling for photocatalytic removal of pollution
in wastewater. Math Comput Appl 22:38–48
5. Razvarz S, Jafari R, Gegov A, Yu W, Paul, S (2018) Neural network approach to solving fully
fuzzy nonlinear systems. In: Fuzzy modeling and control methods application and research.
Nova science Publisher Inc., NewYork, pp 45–68. ISBN 978-1-53613-415-5
6. Razvarz S, Jafari R (2017) Intelligent techniques for photocatalytic removal of pollution in
wastewater. J Electr Eng 5:321–328
7. Jafari R, Razvarz S (2017) Solution of fuzzy differential equations using fuzzy Sumudu
transforms. In: IEEE international conference on innovations in intelligent systems and
applications, pp 84–89
8. Jafari R, Razvarz S, Gegov A, Paul S (2018) Fuzzy modeling for uncertain nonlinear systems
using fuzzy equations and Z-numbers. In: Advances in computational intelligence systems:
contributions presented at the 18th UK workshop on computational intelligence, advances in
intelligent systems and computing, 5–7 September 2018, Nottingham, UK, vol 840. Springer,
Cham, pp 66–107
9. Jafari R, Razvarz S (2018) Solution of fuzzy differential equations using fuzzy Sumudu
transforms. Math Comput Appl 23:1–15
10. Jafari R, Razvarz S, Gegov A (2019) Solving differential equations with Z-numbers by uti-
lizing fuzzy Sumudu transform. In: Arai K, Kapoor S, Bhatia R (eds) Intelligent systems
and applications, IntelliSys 2018, vol 869. Advances in intelligent systems and computing.
Springer, Cham, pp 1125–1138
11. Yu W, Jafari R (2019) Modeling and Control of Uncertain Nonlinear Systems with Fuzzy
Equations and Z-Number. In: IEEE Press Series on Systems Science and Engineering. Wiley-
IEEE Press, Wiley, Hoboken
12. Razvarz S, Jafari R (2018) Experimental study of Al2 O3 nanofluids on the thermal efficiency
of curved heat pipe at different tilt angle. J Nanomater 2018:1–7
13. Razvarz S, Vargas-Jarillo C, Jafari R (2019) Pipeline monitoring architecture based on observ-
ability and controllability analysis. In: IEEE international conference on mechatronics (ICM),
vol 1, pp 420–423
 Nanotechnology Applications in Industry and Heat Transfer 7

14. Razvarz S, Vargas-Jarillo C, Jafari R, Gegov A (2019) Flow control of fluid in pipelines using
PID controller. IEEE Access 7:25673–25680
15. Razvarz S, Jafari R (2016) Experimental study of Al2O3 nanofluids on the thermal effi-
ciency of curved heat pipe at different tilt angle. In: 2nd international congress on technology
engineering and science (ICONTES), Malaysia
16. Jafari R, Yu W, Li X (2016) Solving fuzzy differential equation with Bernstein neural
networks. In: IEEE international conference on systems, man, and cybernetics, Budapest,
Hungary, pp 1245–1250
17. Jafari R, Yu W (2017) Uncertain nonlinear system control with fuzzy differential equations
and Z-numbers. In: 18th IEEE international conference on industrial technology, canada, vol
1, pp 890–895
18. Jafarian A, Measoomynia S, Jafari R (2012) Solving fuzzy equations using neural nets with
a new learning algorithm. J Adv Comput Res 3:33–45
19. Jafari R, Razvarz S, Gegov, A (2019) Neural network approach to solving fuzzy nonlinear
equations using Z-numbers. IEEE Trans Fuzzy Syst 1:1-12
20. Sekhon B (2014) Nanotechnology in agri-food production: an overview. Nanotechnol Sci
Appl 7:31–35
21. Khan MY, Saroj BK, Roy M, Aziz I (2015) A review—emerging use of nano-based carriers
in diagnosis and treatment of cancer novel approaches. Int J Adv Multidiscip Res 2(1):38–54
22. ISO/TR 27628:2007 (2007) Workplace atmospheres - Ultrafine, nanoparticle and nano-
structured aerosols -Inhalation exposure characterization and assessment
23. Stone V, Nowack B, Baun A, van den Brink N, von der Kammer F (2010) Nanomaterials for
environmental studies: classification, reference material issues, and strategies for physico-
chemical characterization. Sci Total Environ 408:1745–1754
24. European Agency for Safety and Health at Work, E-fact, Nanomaterials in maintenance work
(2013) occupational risks and prevention
25. Keshavarz Moraveji M, Razvarz S (2012) Experimental investigation of aluminum oxide
nanofluid on heat pipe thermal performance. Int Commun Heat Mass Trans 39(12):1444–1448
26. ISO/TS 27687:2008(en) (2008) - Nanotechnologies — Terminology and definitions for nano-
objects — Nanoparticle, nanofibre and nanoplate
27. European Commission (2012) Types and uses of nanomaterials, including safety aspects,
Brussels
28. Santos CSC, Gabriel B, Blanchy M, Menes O, García D, Blanco M (2015) Industrial applica-
tions of nanoparticles – A prospective overview, materials today, proceedings, vol 2, pp 456
– 465
29. Park JW (2002) Liposome-based drug delivery in breast cancer treatment. Breast Cancer Res
4(3):95–99
30. Tavman T, Chirtoc T, Schuchmann HP (2008) Experimental investigation of viscosity and
thermal conductivity of suspensions containing nano sized ceramic particles. Arch Mater Sci
Eng 34:99–104
31. Pak BC, Cho YI (1998) Hydrodynamic and heat transfer study of dispersed fluids with
submicron metallic oxide particles. Exp Heat Transf 11:151–170
32. Eastman JA, Choi SUS, Li S, Yu W, Thompson LJ (2001) Anomalously increased effective
thermal conductivities of ethylene glycol-based nanofluids containing copper nanoparticles.
Appl Phys Lett 78(6):718–720
33. Choi SUS, Wang X, Xu W (1999) Thermal conductivity of nano-particle-fluidmixture. J
Thermophys Heat Transf 13(4):474–480
34. Choi SUS (1995) Enhancing thermal conductivity of fluids with nano-particles. ASME
231:99–105
35. Choi SUS (1998) Nanofluid technology: current status and future research. In: Energy, vol 26
8 S. Razvarz et al.

36. American Society of Heating, Refrigerating and Air-Conditioning Engineers (2009): 2009
Ashrae Handbook: fundamentals. ASHRAE, Atlanta
37. Buongiorno J (2005) Convective transport in nanofluids. J Heat Trans 128:240–250
38. Nair R, Varghese SH, Nair BG, Maekawa T, Yoshida Y, Kumar DS (2010) Nanoparticulate
material delivery to plants. Plant Sci 179(3):154–163
39. Phan TD, Debeaufort F, Voilley A (2008) Moisture barrier, wetting and mechanical properties
of shellac/agar or shellac/cassava starch bilayer bio-membrane for food applications. J Membr
Sci 325:277–283
 Applications of Fuzzy Logic, Artificial Neural
Network and Neuro-Fuzzy in Industrial
Engineering

Raheleh Jafari1(B) , Miguel A. Contreras2 , Wen Yu3 , and Alexander Gegov4

1 School of Design, University of Leeds, Leeds LS2 9JT, UK
r.jafari@leeds.ac.uk
2 School of Engineering and Science, Tecnológico de Monterrey, 89600 Altamira,
Tampico, Mexico
macontr@tec.mx
3 Departamento de Control Automático, CINVESTAV-IPN (National Polytechnic Institute),
07360 Mexico City, Mexico
yuw@ctrl.cinvestav.mx
4 School of Computing, University of Portsmouth,
Buckingham Building, Portsmouth PO1 3HE, UK
alexander.gegov@port.ac.uk

Abstract. Artificial intelligence methods namely artificial neural network, fuzzy

logic, and neuro-fuzzy have been effectively utilized in different applications like
business, marketing, control engendering, health care, and social services. To
demonstrate the usage of fuzzy set theory, artificial neural network, as well as
neuro-fuzzy in industrial engineering and also for providing a basis for future
investigation, a literature review of the artificial neural network, fuzzy logic, and
neuro-fuzzy in industrial engineering is conducted in this paper.

Keywords: Fuzzy logic · Artificial neural network · Neuro-fuzzy · Industrial

engineering

1 Introduction

Industrial engineering is one of the fields that artificial neural network, fuzzy logic, and
neuro-fuzzy have found an extensive implementation area. Artificial neural networks
can be considered as the most effective technique over the last few decades that are
extensively utilized in a wide variety of implementations in different fields [1–6]. Arti-
ficial neural networks have been considered as efficient and versatile tools. They have
learning ability as well as model-free characteristics.
Fuzzy systems are appropriate for estimated reasoning, mainly for the system with a
mathematical design which is hard to obtain [7–11]. Fuzzy set theory can be considered
as a major problem designing as well as a solution approach. The main contribution
of fuzzy logic is its ability to present vague data. Fuzzy logic has been implemented

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 9–14, 2020.
https://doi.org/10.1007/978-3-030-45402-9_2
10 R. Jafari et al.

to design systems, which are difficult to define accurately. In recent years, successful
implementations of fuzzy logic in industrial engineering have been reported. Industrial
engineers encounter with numerous problems having incomplete as well as uncertain
information [12–15]. Fuzzy logic theory can be considered as a capable tool for solving
these kinds of problems. Fuzzy logic presents an effective tool to facilitate investigation
in industrial engineering when the dynamics of the decision environment restrict the
accurate evaluation of model parameters.
Neuro-fuzzy has been used in an extensive range of domains [16–18]. It is the combi-
nation of artificial neural networks and fuzzy logic. The cause of utilizing artificial neural
networks with fuzzy logic is that artificial neural networks never make a presumption
on the probability distribution functions of data [19]. Generally, the neuro-fuzzy model
explains solutions better than artificial neural networks [20].
This paper represents details of the application of artificial neural network, fuzzy
logic, and neuro-fuzzy in industrial engineering. In this paper, the most recent researches
in the field of artificial neural network, fuzzy logic, and neuro- fuzzy are covered. Since
some industries have successfully used these techniques, detailed discussions are sup-
plied to stimulate future investigations. This article remaining sections are organized
into four Sections. In Sect. 2 the applications of fuzzy logic in industrial engineering are
given. In Sect. 3 the applications of artificial neural network in industrial engineering
are given. Section 4 presents the applications of neuro-fuzzy in industrial engineering.
Conclusions are included in Sect. 5.

2 Applications of Fuzzy Logic in Industrial Engineering

Fuzzy logic systems have been effectively applied in different industrial fields like auto-
mobile speed control [21], robot arm control [22], water quality control [23], and auto-
matic train operation systems [24]. Fuzzy logic can be utilized for improving the effi-
ciency of the system. Even though fuzzy logic is considered as an approach for presenting
inaccurate and vague information, in [25] the Universal Approximation Theorem states
that the fuzzy logic system can uniformly approximate any nonlinear function to any
degree of preciseness.
The fuzzy inference system utilizes fuzzy set theory for mapping inputs to outputs.
The common fuzzy inference system is Mamdani and Sugeno type. Fuzzy inference
system consists of a fuzzification interface, a rule database, a decision making unit as
well as defuzzification interface, see Fig. 1.
Fuzzy logic systems are extremely effective in highly complicated and nonlinear
processes, and also in the lack of any simple mathematical model. In [26], fuzzy set theory
is utilized for generating a novel pulse discriminator in electric dis- charge machining
procedure. In electric discharge machining procedure, cutting efficiency indexes like
material elimination rate as well as surface harshness are in direct relation with the
electric discharge machining discharge pulses.
In [27], the fuzzy logic is utilized for control of the fluid catalytic cracking unit. In
that paper, the fuzzy logic control as a control method is efficiently applied for improved
procedure control of fluid catalytic cracking in the refinery process industry.
 Applications of Fuzzy Logic, Artificial Neural Network and Neuro-Fuzzy 11

Fig. 1. The flowchart of a fuzzy inference system.

3 Applications of Artificial Neural Network in Industrial

Engineering
In contrast with the fuzzy system, the neural network is desirable at predicting. This
potential originates from the learning ability as well as model-free characteristics. The
design of the artificial neural network is based on a group of interconnected artificial
neurons along with linear or nonlinear transfer functions. Neurons are placed in various
layers such as input layer, hidden layer, as well as output layer. The artificial neural
network learns the relevance between input and output of the system employing an
iterative procedure named training. For every input into the neuron, there is a weight.
Weight is a tune able number that is defined while training the network. A simple model
of the artificial neural network is shown in Fig. 2.
One of the applications of the artificial neural network is in condition monitoring.
Artificial neural network provides a reliable procedure to condition monitoring. Condi-
tion monitoring can be considered as an effective device in maintenance planning, also
it can be utilized for avoiding the unpredicted failures. In [28], artificial neural network
technique is used for experimentally recognizing gears as well as bearings faults related
to a typical gearbox system.
The artificial neural network has been used for evaluating vibrations as well as
recognizing fault existence. Fault detection is considered as a crucial issue for gas turbine
owners to move from preventative repair to predictive repair and hence to decrease the
repair expenses. In [29], artificial neural network technique is used for fault detection of
an industrial gas turbine.
12 R. Jafari et al.

Fig. 2. A simple model of the artificial neural network.

4 Applications of Neuro-Fuzzy in Industrial Engineering

Artificial neural network and fuzzy logic are both considered as model-free numerical
techniques. Each technique uses an uncomplicated algorithmic procedure instead of
a complex mathematical analysis, and also the parameters are tune-able [30]. These
resemblances make it possible to combine the two techniques.
Neuro-fuzzy system is a multi-layer feedforward adaptive network which figures
out the fundamental elements as well as functions of a standard fuzzy logic system.
As fuzzy logic systems are proved to be universal approximators, and since neuro-fuzzy
systems are isomorphic to standard fuzzy logic systems concerning their functions hence
neuro-fuzzy systems are likewise universal approximators [31]. For a special method
in neuro-fuzzy, we can refer to the adaptive neuro-fuzzy inference system which is
considered as one of the initial integrated hybrid neuro-fuzzy models [32].
One of the applications of the neuro-fuzzy system is in polymerization systems. Over
the last few decades solubility of gases in polymers has been of interest to chemical
engineers. In [33], a hybrid grid partitioning adaptive neuro-fuzzy inference system is
utilized to predict carbon dioxide solubility in polymers.
The major aim of current manufacturing industries is to manufacture low cost, superb
quality products in less time [34]. The choice of optimal cutting parameters is crucial in
each machining procedure for increasing the quality of machined productions as well as
to decrease the machining expenses. In [35], an adaptive neuro-fuzzy inference system is
introduced to model and predict surface roughness in ball end milling of a die material.
The algorithm developed in that paper presets the cutting parameters for a favorable
level of surface roughness.
 Applications of Fuzzy Logic, Artificial Neural Network and Neuro-Fuzzy 13

5 Conclusion
In this paper, the recent advances of the artificial neural network, fuzzy logic, and neuro-
fuzzy applications in industrial engineering are provided. The artificial neural network,
fuzzy logic, and neuro-fuzzy are the three major computational intelligence methods.
Utilization of these models can be taken into account as a cheap, highly effective, and
more reliable alternative devices. Therefore, these three methods can provide more ability
to problem resolving than other techniques.

References
1. Razvarz S, Jafari R, Yu W, Khalili A (2017) PSO and NN modeling for photocatalytic removal
of pollution in wastewater. In: 14th International Conference on Electrical Engineering,
Computing Science and Automatic Control (CCE) Electrical Engineering, pp 1–6
2. Jafari R, Yu W (2015) Artificial neural network approach for solving strongly degener-
ate parabolic and burgers-fisher equations In: 12th International Conference on Electrical
Engineering, Computing Science and Automatic Control. https://doi.org/10.1109/iceee.2015.
7357914
3. Jafari R, Razvarz S, Gegov A (2018) A new computational method for solving fully fuzzy
nonlinear systems. In: Computational Collective Intelligence, ICCCI 2018. Lecture Notes in
Computer Science, vol 11055. Springer, Cham, pp 503–512
4. Razvarz S, Jafari R (2017) ICA and ANN modeling for photocatalytic removal of pollution
in wastewater. Math Comput Appl 22:38–48
5. Razvarz S, Jafari R, Gegov A, Yu W, Paul S (2018) Neural network approach to solving fully
fuzzy nonlinear systems. In: Fuzzy modeling and control Methods Application and Research,
Nova science publisher Inc., New York, pp 45–68. ISBN: 978-1-53613- 415-5
6. Razvarz S, Jafari R (2017) Intelligent techniques for photocatalytic removal of pollution in
wastewater. J Electr Eng 5:321–328. https://doi.org/10.17265/2328-2223/2017.06.004
7. Jafari R, Razvarz S (2017) solution of fuzzy differential equations using fuzzy Sumudu
transforms. In: IEEE International Conference on Innovations in Intelligent Systems and
Applications, pp 84–89
8. Jafari R, Razvarz S, Gegov A, Paul S (2018) Fuzzy modeling for uncertain nonlinear sys-
tems using fuzzy equations and Z-numbers. In: Advances in Computational Intelligence
Systems: Contributions Presented at the 18th UK Workshop on Computational Intelligence,
5–7 September 2018, Nottingham, UK. Advances in Intelligent Systems and Computing, vol
840. Springer, Cham, pp 66-107
9. Jafari R, Razvarz S (2018) Solution of fuzzy differential equations using fuzzy sumudu
transforms. Math Comput Appl 23(1):1–15
10. Jafari R, Razvarz S, Gegov A (2019) Solving differential equations with Z-Numbers by utiliz-
ing fuzzy Sumudu transform, intelligent systems and applications. IntelliSys 2018. Advances
in Intelligent Systems and Computing, vol 869. Springer, Cham, pp 1125–1138
11. Yu W, Jafari R (2019) Fuzzy modeling and control with fuzzy equations and Z-number,
IEEE Press Series on Systems Science and Engineering, Wiley-IEEE Press. ISBN-13: 978-
1119491552
12. Razvarz S, Jafari R (2018) Experimental study of Al2O3 nanofluids on the thermal efficiency
of curved heat pipe at different tilt angle. J Nanomater 2018:1–7
13. Razvarz S, Vargas-Jarillo C, Jafari R (2019) Pipeline monitoring architecture based on
observability and controllability analysis. In: IEEE International Conference on Mechatronics
(ICM), Ilmenau, Germany, vol 1, pp 420–423. https://doi.org/10.1109/icmech.2019.872287
14 R. Jafari et al.

14. Razvarz S, Vargas-jarillo C, Jafari R, Gegov A (2019) Flow control of fluid in pipelines
using PID controller. IEEE Access 7:25673–25680. https://doi.org/10.1109/ACCESS.2019.
2897992
15. Jafari R, Yu W, Razvarz S, Gegov A (2019) Numerical methods for solving fuzzy equations:
a Survey. In: Fuzzy Sets and Systems. https://doi.org/10.1016/j.fss.2019.11.003. ISSN 0165–
0114
16. Jafari R, Yu W, Li X (2016) Solving fuzzy differential equation with Bernstein neural net-
works. In: IEEE International Conference on Systems, Man, and Cybernetics, Budapest,
Hungary, pp 1245–1250
17. Jafari R, Yu W (2017) Uncertain nonlinear system control with fuzzy differential equations
and Z-numbers. In: 18th IEEE International Conference on Industrial Technology, Canada,
pp 890–895. https://doi.org/10.1109/icit.2017.7915477
18. Jafarian A, Measoomy NS, Jafari R (2012) Solving fuzzy equations using neural nets with a
new learning algorithm. J Adv Comput Res 3:33–45
19. Castellano G, Fanelli AM (2000) Variable selection using neural-network models. Neuro-
computing 31:1–13
20. Babuska R, Verbruggen H (2003) Neuro-fuzzy methods for nonlinear system identification.
Annu Rev Control 27:73–85
21. Murakami S, Maeda M (1985) Automobile speed control system using a fuzzy logic controller.
In: Sugeno M (ed) Industrial Applications of Fuzzy Control. North-Holland, Amsterdam
22. Scharf EM, Mandix NJ (1985) The application of a fuzzy controller to the control of a multi-
degree-of-freedom robot arm. In: Sugeno M (ed) Industrial Applications of Fuzzy Control.
North-Holland, Amsterdam
23. Yagishita O, Itoh O, Sugeno M (1985) Application of fuzzy reasoning to the water purifi-
cation process. In: Sugeno M (ed) Industrial Applications of Fuzzy Control. North-Holland,
Amsterdam
24. Yasunobu S, Miyamoto S (1985) Automatic train operation system predictive fuzzy control.
In: Sugeno M (ed) Industrial Applications of Fuzzy Control. North-Holland, Amsterdam
25. Wang LX (1994) Adaptive Fuzzy Systems and Control Design and Stability Analysis. Prentice
Hall, NJ
26. Tarng YS, Tseng CM, Chung LK (1997) A fuzzy pulse discriminating system for electrical
discharge machining. Int J Mach Tools Manuf 37:511–522
27. Taskin H, Cemalettin K, Uygun O, Arslankaya S (2006) Fuzzy logic control of a fluid catalytic
cracking unit (FCCU) to improve dynamic performance. Comput Chem Eng 30:850–863
28. Rafiee J, Arvani F, Harifi A, Sadeghi MH (2006) Intelligent condition monitoring of a gearbox
using artificial neural network. Mech Syst Signal Process 21(4):1746–1754
29. Arriagada J, Genrup M, Assadi M, Loberg A (2003) Fault Diagnosis System for an Industrial
Gas Turbine by Means of Neural Networks. In: Proceedings of International Gas Turbine
Congress, Tokyo, Japan
30. Bersini H, Nordvik J, Bonarini A (1993) A simple direct adaptive fuzzy controller derived
from its neural equivalent. In: Proceedings 1993 IEEE International Conference on Fuzzy
System, vol 1, pp 345–350
31. Wang LX, Mendel JM (1992) Fuzzy basis functions, universal approximation, and orthogonal
least squares learning. IEEE Trans Neural Netw 3:807–814
32. Jang J (1993) ANFIS: adaptive network-based fuzzy inference systems. IEEE Trans Syst Man
Cybern 23:665–685
33. Khajeh A, Modarress H, Rezaee B (2009) Application of adaptive neuro-fuzzy inference
system for solubility prediction of carbon dioxide in polymers. Expert Syst Appl 36:5728–
5732
34. Yazdi MRS, Khorram A (2010) Modeling and optimization of milling process by using RSM
and ANN methods. Int J Eng Technol (IJET) IACSIT 2:474–480
35. Hossain MSJ, Ahmad N (2014) A neuro-fuzzy approach to select cutting parameters for
commercial die manufacturing. Proc Eng 90:753–759
 Foundation of Z-Numbers and Engineering
Applications

Sina Razvarz1(B) , Felipe Hernández-Rodríguez2 , Raheleh Jafari3 ,

and Alexander Gegov4
1 Departamento de Control Automático, CINVESTAV-IPN (National Polytechnic Institute),
07360 Mexico City, Mexico
srazvarz@yahoo.com
2 School of Engineering and Science, Tecnológico de Monterrey, Prol. Juan de la Barrera 1241,
25270 Saltillo, Mexico
Felipe.hdz@tec.mx
3 School of Design, University of Leeds, Leeds LS2 9JT, UK
r.jafari@leeds.ac.uk
4 School of Computing, University of Portsmouth,
Buckingham Building, Portsmouth PO1 3HE, UK
alexander.gegov@port.ac.uk

Abstract. Decision-making problems in economics, engineering, risk assess-

ment, forecasting, manufacturing, design and process analysis are naturally deter-
mined by vagueness as well as partial reliability. Generating a formalism for
dealing with Z-numbers is essential to handle these kinds of information. In this
paper, the applications of Z-numbers in engineering are studied. The most current
researches in the area of Z-numbers are covered in this work. Since some engi-
neers have successfully applied Z-numbers, detailed discussions are supplied to
stimulate future investigations.

Keywords: Z-number · Decision-making · Reliability · Engineering applications

1 Introduction
Mainly for the system with a mathematical design which is hard to obtain [1–13]. Fuzzy
set theory is considered as an efficient technique to deal with uncertainty. It has been
effectively implemented to resolve multi-criteria decision making problems. Neverthe-
less, the main issue in decision-making is the reliability of the information associated
with the procedure.
Real-world information is often vague and also is normally represented in natural
language. Moreover, this information is usually partially reliable and a degree of reliabil-
ity is likewise expressed in natural language. To establish a basis to handle this vagueness
and partial reliability, in [14] the idea of Z-number is developed. The Z-number is defined
as an ordered pair Z = (U, V ) of fuzzy numbers utilized to explain a value of a random
variable X , such that U is taken to be a fuzzy constraint on values of X , also V is taken
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 15–24, 2020.
https://doi.org/10.1007/978-3-030-45402-9_3
16 S. Razvarz et al.

to be a reliability which is a value of probability rate of U . Z-numbers are extensively

utilized in a wide variety of implementations in different fields [15–22]. In [23] a novel
analytic hierarchy process technique is proposed based on the Z-numbers for handling
the linguistic decision-making problems. In [24] an arithmetic technique to discrete
Z-numbers is suggested. In [25] an arithmetic technique to continuous Z-numbers is
presented.
To demonstrate the usage of Z-numbers in engineering and also for providing a basis
for future investigation, a literature review of the Z-numbers in engineering is conducted
in this paper. This article remaining sections are organized into three Sections. Section 2
presents some basic definitions related to Z-numbers. In Sect. 3 the applications of
Z-number in engineering are given. Conclusions are included in Sect. 4.

2 Mathematical Preliminaries

We consider the following definitions.

Definition 1. Suppose c is:

1) normal, there is ς0 ∈  where c(ς0 ) = 1,

2) convex, c(βς + (1 − β)ς) ≥ min{c(ς ), c()}, ∀ς,  ∈ , ∀β ∈ [0, 1],
3) upper semi-continuous on , c(ς ) ≤ c(ς0 ) + ε, ∀ς ∈ N (ς0 ), ∀ς0 ∈ , ∀ε >
0, N (ς0 ) is a neighborhood,
4) c+ = {ς ∈ , , c(ς ) > 0} is compact, so c is a fuzzy variable, c ∈ E :  → [0, 1].

The fuzzy variable c is demonstrated as

 
c = c, c̄ (1)

in which c is the lower-bound variable also, c is the upper-bound variable.

Definition 2. The Z-number is made up of two components Z = [c(ς ), p]. The first
component c(ς ) is the restriction on a real-valued uncertain variable ς . The second
component p is a measure of the reliability of c. p can be reliability, strength of belief,
probability or possibility. The Z-number can be stated as Z + -number, in a case c(ς ) be a
fuzzy number also p be the probability distribution of ς . If c(ς ), as well as p, are fuzzy
numbers, the Z-number can be stated as Z − -number.
The Z + -number contains more information when compared with the Z − -number. In
this paper, the definition of Z + -number is utilized, i.e., Z = [c, p], c is a fuzzy number
also, p is a probability distribution.
The most common membership functions which define the fuzzy numbers are the
triangular function

ς−s
μc = H (s, u, v) = v−ς u−s s ≤ ς ≤ u other wise μ = 0 (2)
c
v−u u ≤ ς ≤ v
 Foundation of Z-Numbers and Engineering Applications 17

and trapezoidal function

⎧
⎪ ς−s
⎨≤ς ≤u
u−s s
w−ς
μc = H (s, u, v, w) = ≤ ς ≤ w other wise μc = 0
w−v v
(3)
⎪
⎩ 1u≤ς ≤v

If c represents a fuzzy event in , the real line, then the probability measure can be stated
as

P(c) = μc (ς ) p(ς )dς (4)



where p is the probability density of ς . For discrete Z-numbers we have

n
P(c) = μc (ςl ) p(ςl ) (5)
l=1

Definition 4. The α-level of the Z-number Z = (c, p) is defined as follows

 
[Z ]α = [c]α , [ p]α (6)

where 0 < α ≤ 1. [ p]α is computed by the Nguyen’s theorem

   
[ p]α = p [c]α = p cα , c̄α = P α , P̄ α (7)

where p([c]α ) = { p(ς )|ς ∈ [c]α }. Therefore, [Z ]α is stated as

     
[Z ]α = Z α , Z̄ α = cα , P α , c̄α , P̄ α (8)
 
where P α = cα p ς α , P̄ α = c̄α p(ς̄ α ), [ςl ]α = ς α , ςl α .

The same as with the fuzzy numbers [24], three main operations are defined for the
Z-numbers: ⊕, and , which indicate sum, subtract and multiply respectively.
Suppose Z 1 = (c1 , p1 ) as well as Z 2 = (c2 , p2 ) are two discrete Z-numbers
n 
n
expressing the uncertain variables ς1 and ς2 , so, l=1 p1 (ς1l ) = 1, p2 (ς2l ) = 1.
l=1
The following operation is defined

Z 12 = Z 1 ∗ Z 2 = (c1 ∗ c2 , p1 ∗ p2 ) (9)

Where ∗ ∈ {⊕, , }.
α , cα and [c ]α = cα , cα
The operations used for the fuzzy numbers [c1 ]α = c11 12 2 21 22
are stated as [26],

[c1 ⊕ c2 ]α = [c1 ]α + [c2 ]α = c11

α α
+ c21 α
, c12 α
+ c22
[c1 c2 ]α = [c1 ]α − [c2 ]α = c11
α α
− c21 α
, c12 α
− c22
18 S. Razvarz et al.
  α α α α α α α α 
α min c11 c21 , c11 c22 , c12 c21 , c12 c22 
[c1 c2 ] = α cα , cα cα , cα cα , cα cα (10)
max c11 21 11 22 12 21 12 22
For the discrete probability distributions, the following relation is defined for all p1 ∗ p2
operations
 
p1 ∗ p2 = p1 (ς1 , ι) p2 ς2,(n−ι) = p12 (ς ) (11)
l
The Hukuhara difference is defined as [17],
Z 1 H Z 2 = Z 12
(12)
Z1 = Z2 ⊕ Z12
In a case that Z 1 H Z 2 exists, the α-level can be defined as
α α
[Z 1 H Z 2 ]α = [Zα1 − Zα2 , Z1 − Z2 ] (13)
α Clearly, Z 1 H Z 1 = 0, Z 1 Z 1 = 0.
Moreover, the generalized Hukuhara difference is defined as [22],

1) Z1 = Z2 ⊕ Z12
Z 1 g H Z 2 = Z12 ⇔ (14)
2) Z2 = Z1 ⊕ (−1)Z12
α
By taking into consideration the α-level, we have [Z 1 H Z 2 ]α = [min{Zα1 − Zα2 , Z1 −
α α α
Z2 }, max{Zα1 − Z α2 , Z 1 − Z 2 }] and If also, if Z 1 H Z 2 exists, Z 1 H Z 2 = Z 1 g H Z 2 .
The conditions for the existence of Z 12 = Z 1 g H Z 2 ∈ E are
 α α α
Z α12 = Z α1 − Zα2 and Z 12 = Z 1 − Z 2
1) α α
with Z α12 increasing, Z 12 decreasing, Z α12 ≤ Z 12
 α α α α α α (15)
Z 12 = Z 1 − Z 2 and Z 12 = Z 1 − Z 2
2) α α
with Z α12 increasing, Z 12 decreasing, Z α12 ≤ Z 12
where ∀α ∈ [0, 1].
Suppose c is a triangular function, the absolute value of the Z-number Z = (c, p) is
defined as
|Z(ς )| = (|s1 | + |u1 | + |v1 |, p(|s2 | + |u2 | + |v2 |)) (16)
Let c1 as well as c2 are triangular functions, the supremum metric for Z-numbers Z1 =
(c1 , p1 ) and Z2 = (c2 , p2 ) is expressed as
D(Z1 , Z2 ) = d(c1 , c2 ) + d(p1 , p2 ) (17)
where d(·, ·) is the supremum metrics for fuzzy sets [22]. D(Z1 , Z2 ) has the below
mentioned properties,
D(Z1 + Z, Z2 + Z) = D(Z1 , Z2 )
D(Z2 , Z1 ) = D(Z1 , Z2 )
(18)
D(bZ1 , kZ2 ) = |b| D(Z1 , Z2 )
D(Z1 , Z2 ) ≤ D(Z1 , Z) + D(Z, Z2 )
where b ∈ , Z = (c, p) is Z-number, also c is triangle function.
 Foundation of Z-Numbers and Engineering Applications 19


Definition 5. Suppose Z is the space of Z-numbers. The α−level of the Z-number valued

function H : [0, s] → Z is defined as

H(c, α) = [H(c, α), H(c, α)] (19)


where c ∈ Z, for each α ∈ [0, 1].
Using the definition of Generalized Hukuhara difference, the gH-derivative of H at
c0 is defined as
d 1
H(c0 ) = lim [H(c0 + ζ) gH H(c0 )] (20)
dt ζ→0 ζ

In (27), H(c0 + ζ) as well as H(c0 ) represent symmetric pattern with Z1 and Z2

respectively given in (19).
In this book chapter, the aim of the modeling is to find the Z-number coefficient aι
for the below fuzzy equation
γ
w∗r = aι fι (ϑr ) (21)
ι=1

in such a way that the output of the plant wr approaches to the desired output w∗r .

Definition 6. The distance between two Z-numbers, ψ, φ ∈ Z , is described as the
Hausdorff metric DH [ψ, φ],

DH [ψ, φ] = max{sup(a1 , c1 )∈ψ inf (a2 , c2 )∈φ (d(a1 , a2 )

+ d(c1 , c2 )), sup(a1 , c1 )∈φ inf (a2 , c2 )∈ψ (d(a1 , a2 ) + d(c1 , c2 )} (22)

d(a, c) is the supremum metrics defined for fuzzy sets.


Lemma 1. Let ⊂ Z is a compact set, in this case, is uniformly support-bounded,
i.e. there exists a compact set ⊂ , in such a way that ∀ψ ∈ ,

Supp(ψ) ⊂ . (23)

Lemma 2. Suppose ψ, φ ∈ Z, also α ∈ (0, 1], in this case the following is concluded:

(i) if f, g :  →  are continuous, [f(ψ)]α = f([ψα ]) holds;

(ii) if f :  →  are continuous, so f(Supp(ψ)) = Supp( f (ψ)).

Proof. As (i) concludes from (Yang and Gao 1993), so just (ii) will be proved. Initially,
f(D1 ) = f(D1 ) is shown for D1 ⊂ . As f(D1 ) ⊂ f(D1 ), also f(D1 ) is closed by the
continuity of f, therefore, f(D1 ) ⊂ f(D1 ). Also, for randomly given θ ∈ f(D1 ), there
exists a sequence {an |n ∈ N} ⊂ , a ∈ , in such a way that an → a (n → +∞),
θ = f(a). Since f is continues so, limn→+∞ f(an ) = f(a) = θ.
20 S. Razvarz et al.

However, f(an ) ∈ f(D1 ), therefore, θ ∈ f(D1 ). So f(D1 ) ⊂ f(D1 ). Hence f(D1 ) =

f(D1 ).
Since,

Supp(f(ψ)) = {θ ∈  | (f(ψ))(θ) > 0}

 (24)
f(Supp(ψ)) = f {a ∈  | ψ(a) > 0}

the following is concluded

f(Supp(ψ)) = f({a ∈  | ψ(a) > 0})

= {f(a) ∈  | ψ(a) > 0} (25)

As, {θ ∈  | (f(ψ))(θ) > 0} = {f(a) | ψ(a) > 0}, so

Supp(f(ψ)) = f(Supp(ψ)) (26)

Lemma 3. Suppose G ⊂  is a compact set, also f1 as well as f2 are continuous on

G, ζ > 0, in addition

|f1 (a) − f2 (a)| < ζ, ∀a ∈ G (27)

Hence |supa∈G1 f1 (a) − supa∈G1 f2 (a)| < ζ is valid for each compact set G1 ⊂ G.

Theorem 1. Suppose f :  →  is a continuous function, so for each compact set

 
⊂ Z0 (the set of all the bounded Z-number set) also χ > 0, there exists aι ∈ Z0 ,
ι = 1, 2, . . . , γ, in such a manner that
γ
d(f(ϑ̃), ι=1 fι (ϑ)aι ) < χ ∀ϑ ∈ , ∀ϑ̃ ∈  (28)

in which χ is a finite number.


Theorem 2. If f :  →  is a continuous function, so for each compact set ⊂ Z0 ,

τ > 0 also arbitrary ε > 0, there exists aι ∈ Z0 , ι = 1, 2, . . . , γ, in such a way that
γ
d(f(ϑ), fι (ϑ)aι ) < τ, ∀ϑ ∈ (29)
ι=1

in which, τ is termed as a finite number. The lower as well as the upper limits of the
α-level set of Z-number function approach to τ, however the center approaches to ε.

Theorem 3. Suppose ⊂ Z0 is compact, also is the corresponding compact set of
and f, f̂ :  →  are the continuous functions that,

|f(ϑ) − f̂(ϑ)| < ζ, ∀ϑ ∈ (30)

where ζ > 0. So, ∀ψ1 ∈ , d(f(ψ1 ) − f̂(ψ1 )) ≤ ζ.

 Foundation of Z-Numbers and Engineering Applications 21

3 Applications of Z-Number in Engineering

Multi-sensor data fusion, a functional simulation of the process related to the decision-
making executed by the human brain, has obtained a high growth level over the past few
years across industries. In [27] a new technique with the combination of the Z-number
and Dempster Shafer evidence theory is suggested for modeling as well as processing the
uncertainties in a sensor data fusion system. In that technique, the Z-number is utilized
for modeling the fuzziness as well as the reliability of the sensor data also the Dempster-
Shafer evidence theory is utilized for fusing the uncertain information of Z-numbers. The
major benefits of the suggested technique in [27] are that it produces a perfect robust
degree of reliability to the sensor data, also the supplementary information of multi-
sensors decreases the uncertainty of the fault detection, hence increasing the reliability
of fault identification. The process of fault detection with the technique proposed in [27]
is shown in Fig. 1.

Fig. 1. The process of fault detection.

In [28] Bernstein neural network is used for finding the Z-number solutions of fuzzy
differential equations. In [28] a generalized model of a tank system is proposed which is
in the form of the fuzzy differential equation, see Fig. 2. The tank model is described as
d 1 k
y(t) = − y(t) + (31)
dt CJ C
where K is inflow disturbances of the tank that generates vibration in liquid level y, J is
the flow obstruction which can be curbed utilizing the valve, also C is the cross-section
of the tank.
22 S. Razvarz et al.

Fig. 2. Liquid tank system.

The project management has become a significant approach nearly in all fields of
human activity. Every project is temporary by nature and also it has the beginning and
finish dates. In [29], the capabilities of the Z-numbers in the improvement of the state
of risk evaluation is studied.
In [30], fuzzy equations with Z-numbers are used to design the fuzzy controller. In
[30], a heat source by insulating materials is proposed which is in the form of the fuzzy
equation, see Fig. 3. The insulating materials center is considered to be the source of
heat. The model is described as
A B C D
⊕ = ⊕ ⊕R (32)
K1 K2 K3 K4

Fig. 3. Heat source by insulating materials.

The material widths are not precise and they are in the form of Z-numbers.

4 Conclusion
Engineers tend to deal with various types of measures as well as evaluations. However,
all evaluation of information may not be presented as a clear number.
Z-number has considerable ability in explaining the vagueness of human knowledge.
In this paper, the applications of Z-numbers in engineering are studied. The most current
 Foundation of Z-Numbers and Engineering Applications 23

researches in the area of Z-numbers are covered in this work. Since some engineers have
successfully applied Z-numbers, detailed discussions are supplied to stimulate future
investigations.

References
1. Jafari R, Razvarz S, Gegov A (2018) A new computational method for solving fully fuzzy
nonlinear systems. In: Computational collective intelligence, ICCCI 2018. Lecture notes in
computer science, vol 11055. Springer, Cham, pp 503–512
2. Razvarz S, Jafari R, Gegov A, Yu W, Paul S (2018) Neural network approach to solving fully
fuzzy nonlinear systems. In: Fuzzy modeling and control methods application and research.
Nova Science Publisher, Inc., New York, pp 45–68. ISBN 978-1-53613-415-5
3. Jafari R, Razvarz S (2017) Solution of fuzzy differential equations using fuzzy Sumudu
transforms. In: IEEE international conference on innovations in intelligent systems and
applications, pp 84–89
4. Jafari R, Razvarz S (2018) Solution of fuzzy differential equations using fuzzy Sumudu
transforms. Math Comput Appl 1–15
5. Negoita CV, Ralescu DA (1975) Applications of fuzzy sets to systems analysis. Wiley, New
York
6. Zadeh LA (1968) Probability measures of fuzzy events. J Math Anal Appl 23:421–427
7. Zadeh LA (1975) Calculus of fuzzy restrictions. In: Zadeh LA, Fu KS, Tanaka K, Shimura M
(eds) Fuzzy sets and their applications to cognitive and decision processes. Academic Press,
New York, pp 1–39
8. Zadeh LA (1996) Fuzzy logic and the calculi of fuzzy rules and fuzzy graphs. Multi Val Logic
1:1–38
9. Jafari R, Yu W, Li X (2016) Solving fuzzy differential equation with Bernstein neural
networks. In: IEEE international conference on systems, man, and cybernetics, Budapest,
Hungary, pp 1245–1250
10. Jafari R, Yu W (2017) Uncertain nonlinear system control with fuzzy differential equations
and Z-numbers. In: 18th IEEE international conference on industrial technology, Canada, pp
890–895. https://doi.org/10.1109/icit.2017.7915477
11. Jafarian A, Measoomy Nia S, Jafari R (2012) Solving fuzzy equations using neural nets with
a new learning algorithm. J Adv Comput Res 3:33–45
12. Jafarian A, Jafari R (2019) A new computational method for solving fully fuzzy nonlinear
matrix equations. Int J Fuzzy Comput Model 2:275–285
13. Jafari, R., Razvarz, S., Gegov, A (2018) A novel technique to solve fully fuzzy nonlinear
matrix equations. In: 13th international conference on applications of fuzzy systems and soft
computing, Warsaw, Poland. Advances in intelligent systems and computing. Springer
14. Zadeh LA (2011) A note on Z-numbers. Inform Sci 181:2923–2932
15. Jafari R, Razvarz S, Gegov A, Paul S (2018) Fuzzy modeling for uncertain nonlinear systems
using fuzzy equations and Z-numbers. In: Advances in computational intelligence systems:
contributions presented at the 18th UK workshop on computational intelligence, Nottingham,
UK, 5–7 September 2018. Advances in intelligent systems and computing, vol 840. Springer,
pp 66–107
16. Jafari R, Razvarz S, Gegov A (2019) Solving differential equations with Z-numbers by uti-
lizing fuzzy Sumudu transform. In: Intelligent systems and applications, IntelliSys 2018.
Advances in Intelligent Systems and Computing, vol 869. Springer, Cham, pp 1125–1138
17. Yu W, Jafari R (2019) Fuzzy modeling and control with fuzzy equations and Z-number. IEEE
Press series on systems science and engineering. Wiley-IEEE Press, Piscataway. ISBN-13:
978–1119491552
24 S. Razvarz et al.

18. Jiang W, Xie C, Luo Y, Tang Y (2017) Ranking Z-numbers with an improved ranking method
for generalized fuzzy numbers. J Intell Fuzzy Syst 32:1931–1943
19. Yaakob AM, Gegov A (2015) Fuzzy rule based approach with Z-numbers for selection of
alternatives using TOPSIS. In: Proceedings of the IEEE international conference on fuzzy
systems, pp 1–8
20. Zamri N, Ahmad F, Rose ANM, Makhtar M (2016) A fuzzy TOPSIS with Z-numbers app-
roach for evaluation on accident at the construction site. In: Proceedings of the international
conference on soft computing and data mining, pp 41–50
21. Razvarz S, Jafari R, Yu W (2018) Numerical solution of fuzzy differential equations with Z-
numbers using Fuzzy Sumudu transforms. Adv Sci Technol Eng Syst J (ASTESJ) 3:66–75.
https://doi.org/10.25046/aj030108
22. Jafari R, Razvarz S, Gegov A, Paul S, Keshtkar S (2018) Fuzzy Sumudu transform approach
to solving fuzzy differential equations with Z-numbers. In: Advanced fuzzy logic approaches
in engineering science. IGI Global. https://doi.org/10.4018/978-1-5225-5709-8.ch002
23. Azadeh A, Saberi M, Atashbar NZ, Chang E, Pazhoheshfar P (2013) Z-AHP: a Z-number
extension of fuzzy analytical hierarchy process. In: Proceedings of the 7th IEEE interna-
tional conference on digital ecosystems and technologies, pp 141–147. https://doi.org/10.
1109/DEST.2013.6611344
24. Aliev RA, Alizadeh AV, Huseynov OH (2015) The arithmetic of discrete Z-numbers. Inform
Sci 290:134–155. https://doi.org/10.1016/j.ins.2014.08.024
25. Aliev RA, Huseynov OH, Zeinalova LM (2016) The arithmetic of continuous Z-numbers.
Inform Sci 373:441–460. https://doi.org/10.1016/j.ins.2016.08.078
26. De Barros LC, Bassanezi RC, Lodwick WA (2017) The extension principle of Zadeh and fuzzy
numbers. In: A first course in fuzzy logic, fuzzy dynamical systems, and biomathematics.
Studies in fuzziness and soft computing, vol 347. Springer, Heidelberg, pp 23–41
27. Jiang W, Xie Ch, Zhuang M, Shou Y, Tang Y (2016) Sensor data fusion with Z-numbers and
its application in fault diagnosis. Sensors 16:1509. https://doi.org/10.3390/s16091509
28. Jafari R, Yu W, Li X, Razvarz S (2017) Numerical solution of fuzzy differential equations
with Z-numbers using Bernstein neural networks. Int J Comput Intell Syst 10:1226–1237
29. Nuriyev AM (2019) application of Z-numbers based approach to project risks assessment.
Eur J Interdiscip Stud 5:67–73
30. Jafari R, Yu W, Li X (2017) Numerical solution of fuzzy equations with Z-numbers using
neural networks. Intell Autom Soft Comput 1–7
 Force Balancing of the 2RRR Planar
Parallel Manipulator via Center of Mass
Acceleration Control Using Fully
Cartesian Coordinates

Mario Acevedo1(B) and Marı́a T. Orvañanos-Guerrero2

1
Facultad de Ingenierı́a, Universidad Panamericana,
Álvaro del Portillo 49, 45010 Zapopan, Jalisco, Mexico
mario.acevedo@up.edu.mx
2
Facultad de Ingenierı́a, Universidad Panamericana,
Josemarı́a Escrivá de Balaguer 101, 20290 Aguascalientes, Aguascalientes, Mexico

Abstract. Shaking force balancing of mechanisms usually is achieved

by an optimal redistribution of the moving masses, which allows the can-
cellation or the reduction of the variable dynamic loads on the manipula-
tor’s frame. This procedure usually leads to an increase in the mass of the
original links, motivating the increment on the driving torques and on
the shaking moment. In this article the concept of center of mass acceler-
ation control, presented previously elsewhere, is resumed this time with
the use of fully Cartesian coordinates. The trajectory of the general cen-
ter of mass of the system, calculated out from the motion of the moving
links, is defined as straight line and characterized with a “bang-bang”
profile. In this way it is possible to reduce the maximum acceleration
of the center of mass thus, reducing the shaking force. The suggested
balancing technique is illustrated through computer simulations.

Keywords: Dynamic balancing · Shaking force · Parallel

Manipulator · Fully Cartesian coordinates · Center of mass control

1 Introduction
Robotic industrial manipulators deal with the problem of vibrations transmitted
to the fixed frame during high-speed motion. These vibrations are due to different
factors, one of them is the unbalanced inertia forces which finally produce the
increment of the shaking force and shaking moment. Thus, a primary objective
in the dynamic balancing of manipulators is to eliminate or reduce the variable
dynamic loads transmitted to the base and surrounding structures.
A traditional method used for the dynamic balancing of manipulators has
been the mass redistribution. In this case the objective is to maintain the total
center of mass of the moving links stationary, see [2,5,8]. In practice this is done
adding a set of counterweights and counter-inertias increasing considerably the
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 25–35, 2020.
https://doi.org/10.1007/978-3-030-45402-9_4
26 M. Acevedo and M. T. Orvañanos-Guerrero

mass. This causes an increment in the overall size of the robot-manipulator and
in the efforts at the joints.
Another alternative also traditionally explored is the use of auxiliary struc-
tures. In the same way the aim is to keep the total center of mass of the mech-
anism static. The structures proposed and generally used are the parallelogram
and the pantograph, [6,7,12]. Although this solution leads to relatively complex
mechanical system increasing manipulator’s sizes.
Recently a new approach has been proposed. It is based on the optimal
control of the robot-manipulator center of mass, [4,11]. The method is based
on planning the displacements of the total center of mass of the moving links.
Its trajectory is defined as a straight line with “bang-bang” motion profile. This
approach allows the reduction of the maximum value of the center of mass accel-
eration. Thus leads to a reduction in the shaking force. Using this method it is
possible to ensure the initial and final positions of the end-effector. In this work
this approach is presented using fully Cartesian coordinates, also known as Nat-
ural Coordinates, [10].
An advantage using fully Cartesian coordinates is that the expressions neces-
sary to calculate the general center of mass of the complete system, the shaking
force, and the shaking moment, can be calculated in a direct way. In the authors
opinion this contributes to make the method based on the optimal control of the
robot-manipulator center of mass easy to be programmed and fastest. A com-
plete explanation of their application to find the balancing conditions of a 2RRR
planar parallel manipulator (five-bar mechanism) along with different practical
solution alternatives have been presented in detail in [1].
In this article fully Cartesian coordinates are used to obtain the expres-
sions to calculate the global center of mass and the shaking force of a 2RRR
Planar Parallel Manipulator (2RRR PPM). These expressions are used for its
force balancing applying the method based on the optimal control of the robot-
manipulator center of mass. Comparative results from computer simulations are
presented.

2 Modeling of a 2RRR-PPM Using Fully Cartesian

Coordinates

In Fig. 1 a generic 2RRR-PPM (five-bar mechanism) is presented. All links con-

sidered rigid. Kinematic revolute joints are identified with points A through E,
and points A and E are fixed to the frame. Link lengths are indicated as li , where
i = 1, 2, 3, 4. In this case the centers of mass of the links can be identified with
the letter gi , respectively.
The idea of modeling using fully Cartesian coordinates is to distribute the
mass, ideally concentrated in the center of masses, to the “basic points”. In this
case this is to distribute the masses to the points A through E, the joints.
Thus the system can be seen as a cloud of points, with their corresponding
equivalent mass, moving. Thanks to this change, it is possible to work exclusively
with Cartesian coordinates. Angular coordinates traditionally used in modeling
 Force Balancing of the 2RRR Planar Parallel Manipulator 27

with Reference-Point coordinates can be avoided, as well as their associated

moments of inertia. This facilitates the dynamic analysis greatly.

Fig. 1. Generic 2RRR Planar Parallel Manipulator.

In this way the manipulator has been modeled using the Cartesian coordi-
nates of the “basic points”, that can be organized in a positions vector q:
 T
q = rA rB rC rD rE (1)
where rj = [xj yj ]T are the position vectors that locate the basic points A
through E with respect to the origin of the reference frame.
The dependent coordinates are related by a set of four kinematic constraints,
Φ(q), obtained from the rigid body condition of the bodies:
⎡ ⎤
(xB − xA )2 + (yB − yA )2 − l12
⎢ (xC − xB )2 + (yC − yB )2 − l22 ⎥
Φ(q) = ⎢ ⎥
⎣(xC − xD )2 + (yC − yD )2 − l32 ⎦ = 0 (2)
(xD − xE ) + (yD − yE ) − l4
2 2 2

The coordinates are absolute and can be used to calculate the position of the
center of mass of the system, as well as the shaking force, directly with respect
to the global reference frame (inertial frame). The key point is the assembly of
the mass matrix.

2.1 Calculation of the Mass Matrix

For the assemblage of the mass matrix of the system it is necessary to calculate
first the mass matrix of the elements.
28 M. Acevedo and M. T. Orvañanos-Guerrero

The way to calculate the (4 × 4) mass matrix of a single body using fully
Cartesian coordinates can be reviewed in [10] and more recently in [1]. The
general form of the mass matrix of the link i is shown in Eq. (3):

M1i M2i
Mi = (3)
M3i M4i
where:
2mi x Ji
mi − l + l2 0
M1i = 2mi x Ji (4)
0 mi − l + l2
mi x
l − l2
Ji
− mli y
M2i = mi y mi x Ji (5)
l l − l2
mi x Ji mi y
M3i = l − l2 l (6)
mi y mi x Ji
− l l − l2
Ji
M4i = l2 0 (7)
0 Jl2i
and mi is the mass of link i, and Ii is the moment of inertia of the link i with
respect to its local reference frame.
All the mass matrices of the four mobile bodies of the manipulator can be
calculated in the same way and their terms reorganized to form the (10 × 10)
mass matrix of the system. Its form is presented in Eq. (8).
⎡ ⎤
M11 M21 0 0 0
⎢M31 M41 + M12 M21 0 0 ⎥
⎢ ⎥
M=⎢ 0
⎣ 0
M32 M42 + M43 M33 0 ⎥ (8)
0 M32 M31 + M44 M34 ⎦
0 0 0 M24 M41

This matrix is symmetric and positive definite.

2.2 Calculation of the Global Center of Mass

Once changing to the use of fully Cartesian coordinates and assembling the
global mass matrix of the system, the global center of mass of the system can
be calculated directly as:

mt gt = BMq (9)
4
where mt = i=1mi , is the total mass or the system, gt is the position vector
of the location of the global center of mass, q is vector of dependent Cartesian
coordinates, and M is mass matrix, already defined by Eqs. (1) and (8) respec-
tively. Matrix B is introduced here to sum x and y components in an automatic
way and is composed by (2 × 2) identity matrices, I2×2 , its form is:
 
B = I2×2 I2×2 I2×2 I2×2 (10)
Substituting terms in Eq. (9), the global center of mass of the manipulator
can be calculated as:
 Force Balancing of the 2RRR Planar Parallel Manipulator 29

⎤
⎡
xB
⎢ yB ⎥
⎢ ⎥
⎡ ⎤ ⎢xC ⎥
a b c d e f g h ⎢ ⎢ ⎥
⎥
mt gt = ⎣ ⎦ ⎢ yC ⎥ (11)
⎢xD ⎥
−b a −d c −f e −h g ⎢ ⎥
⎢ yD ⎥
⎢ ⎥
⎣xE ⎦
yE
where:

m2 x2 m1 x1
a= − + + m2 (12)
l2 l1

m2 y 2 m1 y 1
b= − (13)
l2 l1

m3 x3 m2 x2
c= + (14)
l3 l2

m3 y 3 m2 y 2
d= − − (15)
l3 l2

m4 x4 m3 x3
e= − + m3 (16)
l4 l3

m4 y 4 m3 y 3
f= − (17)
l4 l3

m4 x4
g= m4 − (18)
l4

m4 y 4
h= (19)
l4
where (xi , yi ) with i = 1 . . . 4 are the coordinates of the corresponding center
of mass of body i expressed in the local reference frame, see Fig. 2. Note that
the origin of the reference frame coincides with point A, thus (xA , yA ) = (0, 0).
Additionally note the orientation of the local reference frames presented in the
same Fig. 2.

2.3 Calculation of the Shaking Force

The shaking force is the time derivative of the linear momentum. The linear
momentum can be calculated directly deriving equation (9) with respect to time:

mt ġt = BMq̇ (20)

where q̇ is the vector of velocities, the time derivative of vector q.
Taking into account that the mass matrix (8) is constant, and that the veloc-
ities of point A and of point E are zero, the extended expression for the linear
momentum is:
30 M. Acevedo and M. T. Orvañanos-Guerrero

Fig. 2. Model of the 2RRR Planar Parallel Manipulator, showing the local reference
frames orientation. The centers of mass are considered in a general position.

⎡ ⎤
ẋB
⎡ ⎤ ⎢ ẏB ⎥
a b c d e f ⎢ ⎢ẋC ⎥
⎥
mt ġt = ⎣ ⎦ ⎢ ⎥ (21)
⎢ ẏC ⎥
−b a −d c −f e ⎣ ⎥
⎢
ẋD ⎦
ẏD
The shaking force, fsh , can be calculated as the time derivative of the linear
momentum. Thus the shaking force of the manipulator can be finally expressed
in extended form as:
⎡ ⎤
ẍB
⎡ ⎤ ⎢ ÿB ⎥
a b c d e f ⎢ ⎢ ⎥
⎥
fsh = mt g̈t = ⎣ ⎦ ⎢ẍC ⎥ (22)
⎢ ÿC ⎥
−b a −d c −f e ⎢ ⎣ẍD ⎦
⎥

ẏD
Note that the position of the global center of mass, the linear momentum,
and the shaking force are calculated in a very straightforward way once the
mass matrix has been calculated. The mass matrix of the system is very easy to
formulate once the corresponding mass matrix of a single body is known. The
laborious task is the mass assembly, but this can be automated easily.
 Force Balancing of the 2RRR Planar Parallel Manipulator 31

3 Illustrative Example
To show the application of the equations developed in the previous sections, a
specific 2RRR PPM has been selected. A model with the following geometric
parameters have been used: l1 = l4 = 0.36 m, l2 = l3 = 0.3 m, and xE = 0.48 m.
To locate the centers of mass the following values where used: x1 = x4 = 0.18 m,
x2 = x3 = 0.15 m, and y1 = y2 = y3 = y4 = 0. The masses are: m1 = m4 =
0.18 kg and m2 = m3 = 0.15 kg.

3.1 Given Motion to the End-Effector

The trajectory of the end-effector C is given by a straight line limited between

the initial position Ci , with the coordinates xi = 0.04 m, and yi = 0.45 m, and
the final position Cf , with the coordinates xf = 0.44 m, and yf = 0.45 m. The
straight line is traveled following a cycloidal profile with parameters Δs = 0.4 m
in a period of time T = 0.05 s. This can be expressed as in Eq. (23).

t 1 2πt
s(t) = Δs − sin (23)
T 2π T
The coordinates of points B and D, and the corresponding position of the
global center of mass of the system were obtained solving the inverse kinematics
problem, using Eqs. (2) and (11) simultaneously, by a standard Newton-Raphson
algorithm. Note that these equations are non-linear. The trajectory followed by
the tool, point C, and by the global center of mass (CoM) can be seen in see
Fig. 3.

Fig. 3. Tool trajectory, showing initial and final points of the end-effector (point C)
and of the global center of mass (CoM). Dashed lines correspond to the initial position
of the manipulator. Solid lines correspond to the final position.
32 M. Acevedo and M. T. Orvañanos-Guerrero

The global center of mass describes an arc. Its initial position is at

(0.0983, 0.2539), while its final position is (0.3817, 0.2539).
To solve the velocities of point B, D, and the velocity of the CoM, the lin-
ear momentum equations (21) and the kinematic equations for velocities are
solved simultaneously. This last equations are obtained deriving the kinematic
constraints (2) with respect to time as:

dΦ(q) ∂Φ
= q̇ ≡ Φq q̇ = 0 (24)
dt ∂q
where ∂Φ
∂q is the Jacobian matrix. Note that these are a set of simultaneous linear
equations.
In a similar way, to solve the accelerations of point B, D, and the accelera-
tion of the CoM, the shaking force equations (22) and the kinematic equations
for accelerations are solved simultaneously. This last equations can be obtained
deriving the equations for velocities (24) with respect to time as:

Φq q̈ + Φ̇q q̇ = 0 (25)
where Φ̇q is the time derivative of the Jacobian matrix. Again these are a set of
simultaneous linear equations.

3.2 Center of Mass Acceleration Control

As proposed, a second motion profile is given to the manipulator. A trajectory
to the global center of mass of the moving links is defined as straight line and
expressed with a “bang-bang” motion profile. See Fig. 4.
On one hand using this profile it is possible to ensure only the initial and
the final positions of the end-effector but not a straight line trajectory between
them. On the other hand such a control approach allows the reduction of the
maximal value of the center of mass acceleration, leading to the elimination of
the variation in the shaking force. Additionally the shaking force is reduced.
The “bang-bang” law is based on theoretical considerations. Actuators are
unable to achieve discontinuous efforts. Thus this law is proposed to be modi-
fied by trapezoidal profiles to account for the actuators properties in terms of
admissible effort variations. It has been observed that for reasonable actuator
parameters, the results obtained in the cases of the “bang-bang” and trapezoidal
profiles are very close (less than 1%). See [3,9] and [4] for further information.
The corresponding values for points B, D, and C were obtained solving the
inverse kinematics problem, with Eqs. (2) and (11) simultaneously, by the stan-
dard Newton-Raphson algorithm. Note that these equations are non linear. The
position coordinates of the global CoM are used as the degrees of freedom. The
trajectories followed by the tool, point C, and by the global center of mass (CoM)
can be seen in see Fig. 5.
To solve the velocities of points B, C, and D, the linear momentum equa-
tions (21) and the kinematic equations for velocities (24) are solved simulta-
neously. The velocity coordinates of the global CoM were taken as degrees of
freedom.
 Force Balancing of the 2RRR Planar Parallel Manipulator 33

Fig. 4. Bang-bang profile applied to the motion of the global center of mass of the
manipulator.

In a similar way, to solve the accelerations of points B, C, and D, the shaking

force equations (22) and the kinematic equations for accelerations (25) are solved
simultaneously. The acceleration coordinates of the global CoM were taken as
degrees of freedom.

3.3 Shaking Force Comparison

Once solving the kinematics problems: positions, velocities and accelerations,

simultaneously with the equations for the center of mass location, the linear
momentum and the shaking force, it is possible to calculate in straightforward
way the corresponding shaking force.
The procedure, followed for the case of imposing a straight line trajectory
to the end-effector and for a bang-banng trajectory imposed to the global CoM,
gave the corresponding shaking forces compared in Fig. 6.
Note that the variation in shaking force is eliminated after imposing the
“bang-bang” profile to the motion of the global CoM. Additionally the maximum
shaking force is reduced.
34 M. Acevedo and M. T. Orvañanos-Guerrero

Fig. 5. End-effector and global CoM trajectories.

Fig. 6. Shaking force compared. In both cases magnitude is plotted.

4 Conclusions
A method, originally presented in [3] and [4], based on the center of mass accel-
eration control of the manipulator to eliminate the variation and reduce the
shaking force balanced is resumed here using fully Cartesian coordinates. The
procedure to apply this kind of coordinates to obtain the expressions of the
global center of mass of the system, as well as the shaking force, is presented
for a 2RRR PPM. The expressions are obtained in a direct explicit form. These
expressions are used in computer simulations for the shaking force reduction of
the manipulator.

References
1. Acevedo M (2015) Conditions for dynamic balancing of planar parallel manipu-
lators using natural coordinates and their application. In: 14th world congress in
mechanism and machine science, Taipei, Taiwan, 25–30 October 2015
 Force Balancing of the 2RRR Planar Parallel Manipulator 35

2. Agrawal S, Fattah A (2004) Reactionless space and ground robots: novel design and
concept studies. Mech Mach Theory 39(1):25–40. https://doi.org/10.1016/S0094-
114X(03)00102-2
3. Arakelian V (2016) Design of partially balanced 5R planar manipulators with
reduced center of mass acceleration (RCMA). In: Parenti-Castelli V, Schiehlen W
(eds) ROMANSY21 - Robot design, dynamics and control. ROMANSY21 2016.
CISM international centre for mechanical sciences (courses and lectures), vol 569.
Springer, Cham
4. Briot S, Arakelian V, Le Baron JP (2012) Shaking force minimization of high-
speed robots via center of mass acceleration control. Mech Mach Theory 57:1–12.
https://doi.org/10.1016/j.mechmachtheory.2012.06.006
5. Buganza A, Acevedo M (2011) Dynamic balancing of a 2-DOF 2RR planar parallel
manipulator by optimization. In: 13th world congress in mechanism and machine
science, Guanajuato, Mexico, 19–25 June 2011
6. Fattah A, Agrawal S (2003) Design and modeling of classes of spatial reactionless
manipulators. In: Proceedings of the IEEE international conference on robotics
and automation, ICRA 2003, Taipei, Taiwan, pp. 3225–3230
7. Fattah A, Agrawal S (2005) Design arm simulation of a class of spatial reactionless
manipulators. Robotica 3(1):75–81. https://doi.org/10.1017/S0263574704000670
8. Filaretov V, Vukobratovic M (1993) Static balancing and dynamic decoupling of
the motion of manipulation robots. Mech Mach Theory 3(6):767–783. https://doi.
org/10.1016/0957-4158(93)90062-7
9. Geng J, Arakelian V (2019) Design of partially balanced planar 5R symmetrical
parallel manipulators via an optimal motion planning. In: Uhl T (ed) Advances
in mechanism and machine science. IFToMM WC 2019. Mechanisms and machine
science, vol 73. Springer, Cham
10. de Jalón JG, Bayo E (1994) Kinematic and dynamic simulation of multibody
systems: the real-time challenge. Springer, New York. https://doi.org/10.1007/
978-1-4612-2600-0
11. Nenchev DN (2012) Reaction null space of a multibody system with applications
in robotics. Mech Sci 4:97–112. https://doi.org/10.5194/ms-4-97-2013
12. van der Wijk V, Herder JL (2012) Synthesis method for linkages with center of
mass at invariant link point – pantograph based mechanisms. Mech Mach Theory
48:15–28. https://doi.org/10.1016/j.mechmachtheory.2011.09.007
 Approximation of the Workspace
of a Cable-Driven Parallel Robot with a Movable
Gripper

Larisa Rybak , Elena Gaponenko , and Dmitry Malyshev(B)

Belgorod State Technological University named after V.G. Shukhov,

Belgorod, Russian Federation
rlbgtu@gmail.com

Abstract. The article discusses the non-uniform covering method application to

solve the problem of the workspace determination of a cable-driven parallel robot
with a movable gripper for cargo handling in special conditions. The robot is
a rigidly fixed frame connected by tensioned ropes to a platform containing an
axial movement device. The non-uniform covering method is one of the most
well-known deterministic methods for solving global optimization problems. An
algorithm for approximating the workspace of a cable-driven parallel robot using
interval analysis methods has been synthesized. Visualization of simulation results
is carried out by converting a set of three-dimensional boxes into an STL file.

Keywords: Cable-driven parallel robot · Workspace · Non-uniform covering

method · Interval analysis

1 Introduction
Currently, there is a wide variety of parallel structure mechanisms [1–3]. Of particular
interest are cable-driven parallel robots - a special type of kinematic structure, consisting
mainly of a working body connected to a fixed base platform using cables. Cable-driven
parallel robots use cables instead of extension rods to control the position of the output
link. In these mechanisms, the position of the output link is controlled by changing the
length of the cables. Cables are usually wound around coils attached to the base and
driven by a rotating motor. Cable-driven parallel robots have special advantages, such as
a large workspace, ease of assembly and disassembly, high mobility, high load capacity
and ease of reconfiguration. By controlling cable lengths over a wide range, you can
access a very large workspace from a few tens of centimeters to several tens of meters or
more. The use of cables instead of rigid links further reduces weight because the drives
do not change position and are attached to a fixed base so that the only moving parts are
cables and the output link. As a result, we get a robot with higher speed and maneuver-
ability and increased payload. The cost of a cable-driven parallel robots manufacturing
is significantly lower than conventional manipulators. Cable-driven parallel robots are
easy to install. Such a manipulator can be assembled using several inexpensive winches
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 36–43, 2020.
https://doi.org/10.1007/978-3-030-45402-9_5
 Approximation of the Workspace of a Cable-Driven Parallel Robot 37

and cables. Also, since motors do not need to be installed near a moving platform, they
are suitable for use in hazardous environments. Also, their carrying capacity is relatively
high, even comparable to construction cranes.
Particular attention is paid to the topic of cable-driven parallel robots by J.P. Merlet,
A. Pott, C. Gosselin, J.B. Izard, M. Michelin, S. Baradat, and others. So, Pott and others
developed a family of cable-driven parallel robots called IPAnema that are designed
for industrial processes, in [4] methods for planning and modeling a cable robot are
described in detail. J.P. Merlet and D. Daney developed the light and mobile cable
parallel robot Marionet [5], which is designed for rescue operations in hard-to-reach
places. In [6], dynamic planning of the trajectory of 3-DOF CDPRS is considered. The
proposed approach to planning trajectories can be used to plan dynamic trajectories that
go beyond the static working space of the mechanism, opening up new applications
and opportunities for cable robots. In [7], dynamic analysis and classification of the
workspace were considered based on the general equation of motion of a cable robot and
the one-way property of cables. A qualitative comparison of various types of workspaces.
In [8], based on a structural analysis, the classification of cable robots was compiled and
considered in detail: by the method of placing the drums, by the number of degrees
of freedom of the working body and the number of flexible links, by the method of
application, by the type of base, by the method of cables fixing.
An important characteristic of parallel structure mechanisms is their workspace,
within which a working tool should be located when performing technological oper-
ations. To determine the workspace, it is known to use various methods: geometric,
numerical, and discretization methods. In [9], the method of non-uniform coverings for
approximating the set of solutions of a system of nonlinear inequalities was considered,
and in [10, 11], the application of this method to determine the workspace of some types
of planar robots was considered.

2 Problem Statement
Consider the application of the non-uniform covering method to determine the workspace
of a Cable-driven parallel robots designed to work in special conditions. For this purpose,
we will use the methods of interval analysis described in detail in [12, 13]. The circuit
technical solution of such a robot is shown in Fig. 1. The manipulator consists of four
columns, four cables, which are connected at one end to a moving platform at points
B1 , B2 , B3 , B4 . A gripping device with the possibility of axial movement is fixed on
the platform. The second ends of the cables passed through the pulleys, indicated by
points A1 , A2 , A3 , A4 , mounted on columns, are fixed on the drums D1 , D2 , D3 , D4 ,
respectively. Under the action of the weight of the load mg, fixed at point C, tensile
forces T1 , T2 , T3 , T4 arise in the cables. A change in the position of the load securing
point occurs due to a change in the cable lengths during rotation of the drums by gear
motors M1 , M2 , M3 , M4 . The gripping device is a movable platform connected with
cables. An output link C is located on the mobile platform, which has the possibility of
axial movement of h along the Z axis. This mechanism allows you to move objects in
hard-to-reach places with special conditions without human intervention, for example,
working with radioactive elements, radiation sources, working with elements that can
38 L. Rybak et al.

irradiate cells lead to their mutation. They can cause irreparable harm to health. This
category also includes work at nuclear and thermal power plants.

Fig. 1. Scheme of a cable-driven parallel robot for cargo moving

Despite their many advantages, there are several problems associated with controlling
the movement of robot cables. One of the drawbacks is that the cables can pull, but not
push, this leads to a one-way restriction, in which the cables must always be kept in a
tight state. The provisions in which at least one of the cables is loosened are special and
are not considered in this article.
We introduce restrictions on the cable tension forces T i
   
T i := T i , T i = 0 ≤ T i ≤ Ti ≤ T i (1)

We write down the intervals that describe the ranges of variation of the cosines and sines
of the angles αi and βi :
   
S Ai := S Ai , S Ai = S Ai ≤ sin αi ≤ S Ai , i ∈ 1, . . . , 4.
   
S Bi := S Bi , S Bi = S Bi ≤ sin bi ≤ S Bi , i ∈ 1, . . . , 4.
 Approximation of the Workspace of a Cable-Driven Parallel Robot 39

   
C Ai := C Ai , C Ai = C Ai ≤ cos αi ≤ C Ai , i ∈ 1, . . . , 4.
   
C Bi := C Bi , C Bi = C Bi ≤ cos bi ≤ C Bi , i ∈ 1, . . . , 4.

The tension forces in the flexible links can be calculated by solving systems of nonlinear
equations having the form:
⎧
⎨ T 1 S A1 C B1 + T 2 S A2 S B2 − T 3 S A3 C B3 − T 4 S A4 S B4 = 0
T S S − T 2 S A2 C B2 − T 3 S A3 S B3 + T 4 S A4 C B4 = 0 (2)
⎩ 1 A1 B1
T 1 C A1 + T 2 C A2 + T 3 C A3 + T 4 C A4 − mg = 0

where
z Ai − Z Bi
C Ai = ;
(X Bi − x Ai )2 + (Y Bi − y Ai )2 + (z Ai − Z Bi )2

(X Bi − x Ai )2 + (Y Bi − y Ai )2
S Ai = ;
(X Bi − x Ai )2 + (Y Bi − y Ai )2 + (z Ai − Z Bi )2
Y B1 − y A1
S B1 = ;
(X B1 − x A1 )2 + (Y B1 − y A1 )2
X B2 − x A2
S B2 = ;
(X B2 − x A2 )2 + (y A2 − Y B2 )2
y A3 − Y B3
S B3 = ;
(x A3 − X B3 )2 + (y A3 − Y B3 )2
x A4 − X B4
S B4 = ;
(x A4 − X B4 )2 + (Y B4 − y A4 )2
X B1 − x A1
C B1 = ;
(X B1 − x A1 )2 + (Y B1 − y A1 )2
y A2 − Y B2
C B2 = ;
(X B2 − x A2 )2 + (y A2 − Y B2 )2
x A3 − X B3
C B3 = ;
(x A3 − X B3 )2 + (y A3 − Y B3 )2
Y B4 − y A4
C B4 = .
(x A4 − X B4 )2 + (Y B4 − y A4 )2

Wherein
a
X B1 = X B2 = X C − ,
2
40 L. Rybak et al.

a
X B3 = X B4 = X C + ,
2
a
Y B1 = Y B4 = YC − ,
2
a
Y B2 = Y B3 = YC + ,
2
Z Bi = Z C + H,

where
   
X C := X C , X C = X P ≤ xC ≤ X P
   
Y C := Y C , Y C = Y C ≤ yC ≤ Y C
   
Z C := Z C , Z C = Z C ≤ z C ≤ Z C
   
H := H, H̄ = H ≤ h ≤ H̄

The left sides of the system of Eq. (2) are functions in the form g j , j ∈ 1, . . . , 3:
g1 = T 1 S A1 C B1 + T 2 S A2 S B2 − T 3 S A3 C B3 − T 4 S A4 S B4 , …, g3 = T 1 C A1 +
T 2 C A2 + T 3 C A3 + T 4 C A4 − mg. The above equations allow us to find solutions to the
system of Eq. (2) that determine the limitations of the workspace.

3 Algorithm Synthesis

Taking into account formulas (1–2), an algorithm for approximating the workspace of
a cable-driven parallel robot was synthesized, which is shown in Fig. 2. It works with a
system of equations written in a general form:
⎧
⎪
⎪ g1 (x) = 0,
⎨
...
(3)
⎪
⎪ g (x) =0
⎩ m
ak ≤ xk ≤ bk , k = 1, .., n.

The initial box Q, which covers the entire set of solutions X, is determined by the interval
constraints ak ≤ xk ≤ bk , k = 1, .., n. Consider an arbitrary box B. Let m(B) =
max j=1,..,m minx∈B g j (x) and M(B) = min j=1,..,m maxx∈B g j (x). If m(B) > 0 or
M(B) < 0, then B does not contain possible points for the system (3). If the box
can not be excluded, it is divided into two smaller boxes, if its diameter is greater
than the specified approximation accuracy δ. The algorithm works with three lists of
n-dimensional boxes P (current list), P A (coverage) and P E (excluded boxes). Each of
the dimensions of the boxes corresponds to the intervals X C , Y C , and Z C .
 Approximation of the Workspace of a Cable-Driven Parallel Robot 41

Fig. 2. The block diagram of the algorithm for determining the workspace and optimization of
geometric parameters.

The algorithm works as follows:

1. Set the geometric parameters of the robot, the intervals H, Ti , i ∈ 1, . . . , 4 and
approximation accuracy δ.
2. At the first step of the algorithm, lists P A and P E are empty, list P consists of
only one box Q, including intervals X C , Y C and Z C : X C = [0, x A3 ], Y C = [0, y A3 ],
Z C = [0, z A3 ].
3. Remove the box B from the list P.
4. Calculate m(B) > 0 and M(B).
5. If m(B) > 0 or M(B) < 0, then add B to the list P E and go to step 8.
6. If d(B) < δ, then add B to the list P A and go to step 8.
7. In either cases, B is divided into two equal boxes along the edge with the longest
length. Insert these boxes at the end of the list P, that is, P := P ∪ {B1 } ∪ {B2 }.
8. If P = ∅, then go to step 3.
The algorithm is implemented in the C++ programming language using the Snow-
goose interval analysis library [14]. Visualization of the simulation results is carried out
by converting the list of boxes P I to the universal format of 3D models - the STL file.
42 L. Rybak et al.

4 Simulation Results
A computational experiment was conducted for the following geometric parameters of
the mechanism: H = [0 mm, 200 mm], T = [20 N, 100 N], x A3 = x A4 = y A2 = x A2 =
z Ai = 1000 mm. The simulation results are presented in Fig. 3. The calculation time for
the approximation accuracy δ = 10 mm on a personal computer was 5 min 36 s. The
volume of the workspace is 0.474 m3 .

Fig. 3. The results of the simulation.

Figure 4 shows the dependence of the workspace volume of the cable-driven parallel
robot on the range of the output link C movement along the Z axis.

Fig. 4. Dependence of the workspace volume on the work tool axial movement.
 Approximation of the Workspace of a Cable-Driven Parallel Robot 43

5 Conclusion
Thus, for the proposed cable-driven parallel robot, effective numerical methods and
algorithms were developed and tested that allowed to determine the workspace. The
calculation time for the approximation accuracy δ = 10 mm on a personal computer
was 5 min 36 s. A further area of research is the optimization of the geometric parameters
of the robot to ensure the required workspace.

Acknowledgments. This research has been supported by the RFBR according to the research
project No. 18-31-20060.

References
1. Merlet J-P (2007) Parallel robots, 2nd edn. Springer, Heidelberg
2. Kong H, Gosselin CM (2007) Type synthesis of parallel mechanisms. Springer, Heidelberg
3. Aleshin AK, Glazunov VA, Rashoyan GV, Shai O (2016) Analysis of kinematic screws that
determine the topology of singular zones of parallel-structure robots. J Mach Manuf Reliab
45(4):291–296
4. Pott A, Mütherich H, Kraus W, Schmidt V, Miermeister P, Verl A (2013) IPAnema. A family of
cable-driven parallel robots for industrial applications. In: Bruckmann T, Pott A (eds) Cable-
driven parallel robots. Mechanisms and machine science, vol 12. Springer, Heidelberg, pp
119–134
5. Merlet JP, Daney DA (2010) A portable, modular parallel wire crane for rescue operations.
In: 2010 IEEE international conference on robotics and automation (ICRA), pp 2834–2839
6. Gosselin C (2013) Global planning of dynamically feasible trajectories for three-DOF spatial
cable-suspended parallel robots. In: Bruckmann T, Pott A (eds) Cable-driven parallel robots.
Springer, Heidelberg, pp 3–22
7. Duan QJ, Xuechao D (2014) Workspace classification and quantification calculations of
cable-driven parallel robots. In: Advances in mechanical engineering. Hindawi Publishing
Corporation, pp 1–9
8. Gaponenko EV, Rybak LA, Kholoshevskaya LR (2019) Structural analysis and classification
of robotic systems with driving mechanisms on the basis of cable elements. Bulletin BSTU
named after V.G. Shukhov 9:126–136 (2019)
9. Evtushenko Yu: Numerical methods for finding global extreme (case of a non-uniform mesh).
USSR Comput Maths Math Phys 11: 1390–1403 (1971)
10. Evtushenko Y, Posypkin M, Rybak L, Turkin A (2018) Approximating a solution set of
non-linear inequalities. J Global Optimization 7:129–145
11. Gorchakov AYu, Ignatov AD, Malyshev DI, Posypkin MA (2019) Parallel algorithm for
approximating the working area of the robot. Int J Open Inf Technol 7(1):1–7
12. Jaulin L (2001) Applied interval analysis: with examples in parameter and state estimation,
robust control and robotics. Springer, New York
13. Moore RE (1966) Interval analysis. Prentice-Hall, Englewood Cliffs
14. Posypkin M, Usov A. Basic numerical routines. https://github.com/mposypkin/snowgoose.
Accessed 18 Nov 2019
 Design of a New Game for Teaching
Assembly Process

Tianxiong Wang1 and Rafiq Ahmad2(B)

1 Beijing Institute of Technology, Beijing, China
2 University of Alberta, Edmonton, Canada

rafiq.ahmad@ualberta.ca

Abstract. Assembly practice is regarded as a great way to deal with the train-
ing needs of the increasing number of mechanical companies as students can be
equipped with hands-on skills and working experience from assembly training.
However, the current assembly-related courses with hands-on activities have a
series of problems including cost and safety. This paper analyzes the possibility of
applying serious games to assembly training and introduces a novel serious game
to help mechanical engineering students participate the assembly process in a fun
environment. The game is developed by Unity3D and based on BCN3D MOVEO
robotic arm. Learners can get familiar with the assembly process from the game
and instructors can customize the assembly process for their students to learn. The
benefits this game can have on mechanical engineering education are examined

Keywords: Assembly · Serious game · Gamification

1 Introduction

Learning mechanical engineering requires students to be actively involved in the learning

process. However, a majority of courses which are related to mechanical engineering are
given by lecture-based approach [1]. This traditional teaching method has obvious disad-
vantages because it is not an effective motivator as students can only receive information
from textbooks and instructors without taking active part in education [2]. Furthermore,
the traditional approach leaves students with poor practical abilities as students cannot
get enough opportunities to deal with workplace problems which are complex, ambigu-
ous and ill-structured without a clear goal and a single solution [3], compared with the
clear and direct principles taught by traditional lectures. In order to help students to
get love for learning and prepare them for the real world of practice, project-based and
training-based teaching approaches are supposed to be carried out [4].
Our game is designed to respond to this demand and help more mechanical engineer-
ing learners take part in the assembly process conveniently and economically. Assembly
training is in the core of integrated courses which are related to mechanical engineer-
ing. Through assembly practice, students are able to develop operative skills in real
manufacturing environment [5].

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 44–52, 2020.
https://doi.org/10.1007/978-3-030-45402-9_6
 Design of a New Game for Teaching Assembly Process 45

Our game is designed in Unity, which means that the game is three dimensional,
portable and machine independent. These characteristics make the game a good sup-
porting tool for mechanical engineering instructors. Furthermore, the scoring system
and the game-based design principle enable students to study not only independently
but also cooperatively while also having fun.
Although our design focuses on the assembly process of one machine, our research
can be extended to other assembly simulator design projects in the field of engineering.
The paper is organized as follows. Section 2 introduces related work about education
of assembly in mechanical engineering. The concept of gamification and other related
theoretical background information are explained in Sect. 3. Section 4 briefly introduces
BCN3D MOVEO, which is the machine chosen for our game. The development of the
game will be introduced in Sect. 5. Section 6 provides conclusions.

2 Related Work
Nowadays, many companies and laboratories have already designed games or software-
based simulators to help students learn practical skills in the field of mechanical engi-
neering. NIU-Torcs is a car racing video game designed to help mechanical engineering
undergraduates learn computational techniques and concepts. In the game, a player is
required to write a C++ program to get his or her car to move on the track in the way
prescribed by the game [6]. Although the game can help students learn engineering cal-
culations strategies, it is not able to improve students’ hands-on skills because the game
concentrates on programming rather than assembly.
Mecagenius® is a game designed to teach students mechanical manufacturing and
machining process [7]. But the game is developed by Adobe Flash technology, which
implies that all the components in the game is two dimensional so students cannot have a
complete perspective of all the assemblies in the game. Moreover, the correct operational
procedures in each activity of the game cannot be changed, which means that teachers
are not able to design their own operational procedures for their students.
On the contrary, our game is three dimensional so users can observe every component
of the assembly clearly in our game and the simple user interface makes the game easy-
to-use and easy-to-learn for new learners. In addition, teachers can change the correct
assembly procedure according to their own teaching plans and project assessments.

3 Theoretical Background
3.1 Gamification
The term “gamification” was first documented used in 2008 but there are many parallel
or similar concepts which have already existed before 2008 or continue being introduced
after the appearance of gamification. Although the term originated in 2008, it was not
widely adopted at the beginning. Gamification was a contested term before 2011 with
many researchers using other terms for their similar practice [8].
Deterding [8] proposed a definition of “gamification” as the use of game design
elements in non-game context in 2011, which is regarded as the most acknowledged
46 T. Wang and R. Ahmad

definition so far. In order to reach this definition, Deterding make a distinction between
playing and gaming. Gaming is characterized by explicit rule systems and goals. In
contrast, the main goal of playing is to entertain the users so playing denotes a free-form
and open recombination of behaviors and meanings [9]. The goal of our project is to
apply gamification in the educational context, so the following discussion will focus on
gaming rather than playing.

3.2 Serious Games

The term “serious game” refers to games which are designed to help users develop new
knowledge and skills rather than mere entertainment [7]. Through the application of
computer, users of serious games can experience situations which are impossible for
them to meet due to safety and cost. Nowadays, serious games have already been widely
applied in a large number of fields, e.g. health care, education, government, military.
Serious games can be classified according to their fields of application: America’s
Army, which was released in 2002, is one of the most well-known serious games in mili-
tary. The U.S. military invested the game and the game successfully helped U.S. military
to recruit new volunteer soldiers with 15% of the cost of other recruiting problems. Dance
Revolution is a healthcare serious game, which requires associated hardware. The game
successfully combined game elements with physical activities to make players be more
engaged in physical exercises [10]. E and Eve’s Electrical Endeavors is an example
of educational serious games. The game was developed by the Eindhoven University
of Technology and aims to help players to learn knowledge about electrical engineer-
ing. According to the investigation of the players of the game, students who used the
game to study performed remarkably better than those who received information via
textbooks [11].

3.3 Advantages of Serious Games in Engineering Education

It has been proved that adding game elements in educational context can significantly
improve students’ attention, participation and interests [12, 13]. Moreover, games can
provide students with opportunities to improve communication skills, cooperate with
other students and experiment dangerous conditions in a safe manner. By integrating
game elements in curriculum or offering game-based courses, schools can increase their
students’ learning effects and enthusiasm towards study. [14] provides a study in which
students’ engagement in a mechanical engineering course built around a video/computer
game was measured. The authors get the conclusion that students are eager to learn and
take on different challenges in a game-based learning environment.
Furthermore, the evolution of industries has already increased the demand for
employees with hands-on experience. Although assembly is a good way to improve
students’ hands-on skills and computer-based assembly exercise is able to satisfy this
demand, there are no assembly-oriented serious games available on the market. A serious
game about assembly with the characteristics of efficiency, convenience and fun can fill
the gap of gamified experiences for assembly process and help to cope with the training
needs of industries.
 Design of a New Game for Teaching Assembly Process 47

4 Open Source MOVEO Robot

We chose BCN3D MOVEO [15] as the assembly of our game. The picture of MOVEO
can be seen in Fig. 1. MOVEO is a fully open source 3D printed robotic arm developed by
BCN3D Technologies. The structure of MOVEO is simple and it is therefore suitable for
education. This robotic arm consists of five major parts and is controlled by the software
Arduino. There are total 232 components in MOVEO, including screws, locknuts, printed
parts, etc. As BCN3D Technologies puts it, MOVEO was developed with an open and
education wish.

Fig. 1. MOVEO robotic arm (BCN3D Technologies)

All the necessary information about assembling a MOVEO robotic arm is available
on GitHub [16]. Thanks to BCN3D Technologies and GitHub, one can get the CAD and
STL files of all the components of MOVEO. These files were converted to OBJ files in
SolidWorks and used in Unity3D, the development platform of our game. Furthermore,
BCN3D Technologies also provides the user manual, which gives the instruction of the
assembly process on GitHub. The official user manual is used in the instruction module
of the designed game.

5 Proposed Game
De Paz [17] divided the general guidelines to gamification into three phases, which will
be briefly introduced afterwards.
Phase 1: Preparation. In this phase, designers of are supposed to set up the goals of
their game, be clear about what problems they want to gamify and gain some knowledge
about different game elements which can be involved in their game.
Phase 2: Design. At this stage, the main objective is to determine the basic design
and the use of game elements. As new ideas may come up throughout the project, the
ideas that designers have in this stage can be changed from time to time.
Phase 3: Implementation. The target of this stage is to build the physical system
and put it into play.
Then, we will introduce how we design and implement our game according to the
design principle proposed by De Paz.
48 T. Wang and R. Ahmad

5.1 Preparation
The goal of the design is to provide students with a computer-based system on which
they can assemble a machine, learn some knowledge about assembly process and get
to know parameters of different basic components (screws, motors, bearing and so on).
In addition, instructors are supposed to be authorized to change the assembly sequence
or even the final machine according to their own teaching plans. In order to motivate
students to play the game, some elements of competition are expected to be added into
the game design [18]. A good scoring system can not only help teachers know the
performance of different students but also make students compete with their friends and
improve their skills further.

5.2 Design

Level Design
According to the user manual of MOVEO, the whole assembly can be divided into
five sub-assemblies, so there are five levels in the game. The order of these five levels
corresponds to the complexity of the five sub-assemblies. Level one is the easiest, while
level five is the hardest. Before entering a new level, students should finish all the easier
levels first.

Selection and Sequencing of 3D Printed Components

The assignment of students in these section is to choose the right 3D printed components
and put them in a correct order according to the instruction of their current level. Figure 2
shows the user interface of this section.

Fig. 2. User interface of the section of 3D printed components

Different from basic components, 3D printed components are complex, asymmetric

and irregular, so traditional 2D game development platform, such as Adobe Flash, cannot
be used. That is the reason why we choose Unity3D to develop our assembly game.
 Design of a New Game for Teaching Assembly Process 49

Once students put the mouse on one component, it will rotate so students can observe
the component from all perspectives. This design is to make the game get close to real
assembly environment and make students see the components more clearly in a computer
game.
In order to make novice players play the game easily, the user interface was designed
as concisely as possible. First, the player selects the “Instruction” button to read the
instruction of the sub-assembly in his/her level. Then, he/she can click on the white
square to select the 3D printed parts he/she wants to use in the sub-assembly. After
selection, the player need to input the sequence numbers of the 3D printed parts he/she
has selected in the input field next to the white square. “P1”, “P2”, “P3” and “P4”
buttons at the lower right corner of the interface control the page transition as there
are total thirty-two 3D printed parts in MOVEO but one page can only display eight
of them. The clock at the top left corner displays the time the player has already spent
in his level. The “Enter” button enables players to check whether their selection and
sequencing are correct or not. If it is right, the system will transfer to next module which
will be explained.

Selection of Basic Components

In this section, players are required to choose the right components and input the quantity
according to the instruction of their current level. Figure 3 shows the user interface of
this section.

Fig. 3. User interface of the section of basic components

Players can click on the “Enter” button to check whether their input values are
correct or click on the “Back” button to go back to the welcome window of this section
to choose other basic components. Moreover, players can transfer between the section
of basic components and 3D printed components at any time with all the data saved
automatically.
After these two sections, the scoring system of the game will give out the final score
of the player.
50 T. Wang and R. Ahmad

Scoring System
The final score depends on the total time a player spend and the number of clicks he/she
makes in an assembly process. The score is calculated by the following equation:

Final score = 5000 − Total Time(mins) ∗ 100 − Number of Clicks ∗ 100

Time is a very important parameter to judge a player’s level of proficiency in assem-

bly. The number of clicks is also very important as some players may make many
mistakes and make many changes in a short time or even make random attempts in the
game. Therefore, the scoring system is designed to make players who can finish the
assembly both quickly and accurately to get high scores.

Custom Settings
Instructors who have correct usernames and passwords can get into the custom setting
section in which they can define the required 3D printed components from the thirty-two
components and the assembly sequence according to their preference. Figure 4 shows
the user interface of custom setting.

Fig. 4. User interface of custom setting

For each level, instructors need to input two strings with the length of 32 characters.
The first string represents selection, which means whether a certain component should be
selected in this level. 1 stands for yes, while 0 stands for no. The second string represents
sequencing. Instructors need to set the sequence number for each of the components that
should be selected.

5.3 Implementation
The game can be played in the class as a way to train students’ assembly skills. Instructors
can use the official assembly methods of MOVEO or define their own assembly strategies
and provide students with custom instructions. Schools can offer game-based curriculum
with this game or offer real-assembly courses and use this game as a teaching aid. For
 Design of a New Game for Teaching Assembly Process 51

Students, they can play the game after class to preview or review. The game can be also
used in tests due to the scoring system. For an initial evaluation, the game was tested
by several group members. They expressed their interest and judged that the game is a
good learning tool for assembly practice. We intend to evaluate the performance of the
game in future works.

6 Conclusions

This paper presented a brief research on gamification and serious games and discussed
the value of applying serious games to the education of mechanical engineering. Assem-
bly serious games can greatly improve the academic performance and hands-on skills
of students but there is a lack of assembly-oriented serious games. Then, the paper
introduced the design of an assembly game, which is based on BCN3D MOVEO and
developed by Unity3D. The purpose of this design was to help students learn basic
assembly skills and help schools set up assembly courses. We believe that this game
can be improved in several ways, which will be considered in future work. For example,
the graphics will be upgraded and the custom setting is better not to require users to
set all the five levels at one time. The future work will also focus on validating this game
in engineering courses. In future, the game evaluation will also be extended to school
students.

References
1. Yadav A, Shaver GM, Meckl P (2010) Lessons learned: implementing the case teaching
method in a mechanical engineering course. J Eng Educ 99(1):55–69
2. Hamada M, Sato S (2011) A game-based learning system for theory of computation using
Lego NXT Robot. Procedia Comput Sci 4:1944–1952
3. Jonassen D, Strobel J, Lee CB (2006) Everyday problem solving in engineering: lessons for
engineering educators. J Eng Educ 95(2):139–151
4. Prince MJ, Felder RM (2006) Inductive teaching and learning methods: definitions,
comparisons, and research bases. J Eng Educ 95(2):123–138
5. Ahmad R et al (2018) Alberta learning factory for training reconfigurable assembly process
value stream mapping. Procedia Manuf 23:237–242
6. Coller B (2007) Implementing a video game to teach principles of mechanical engineering.
In: 114th annual ASEE conference and exposition
7. Pons-Lelardeux C et al (2015) Didactic study of a learning game to teach mechanical
engineering. Procedia Eng 132:242–250
8. Deterding S, et al (2011) From game design elements to gamefulness: defining gamification.
In: Proceedings of the 15th international academic MindTrek conference: envisioning future
media environments. ACM
9. Müller BC, Reise C, Seliger G (2015) Gamification in factory management education–a case
study with Lego Mindstorms. Procedia CIRP 26:121–126
10. Michael DR, Chen SL (2005) Serious games: games that educate, train, and inform. Muska
& Lipman/Premier-Trade, Boston (2005)
11. Stege L, Van Lankveld G, Spronck P (2011) Serious games in education. Int J Comput Sci
Sport 10(1):1–9
52 T. Wang and R. Ahmad

12. Barata G et al (2013) Improving participation and learning with gamification. In: Proceedings
of the first international conference on gameful design, research, and applications. ACM
13. Barata G et al (2013) Engaging engineering students with gamification. In: 2013 5th inter-
national conference on games and virtual worlds for serious applications (VS-GAMES).
IEEE
14. Coller BD, Shernoff DJ (2009) Video game-based education in mechanical engineering: a
look at student engagement. Int J Eng Educ 25(2):308
15. BCN3D. BCN3D MOVEO – A fully Open Source 3D printed robot arm. https://www.
bcn3dtechnologies.com/en/bcn3d-moveo-the-future-of-learning/
16. BCN3D. BCN3D-MOVEO: Open Source 3D Printed Robotic Arm for educational purposes.
https://github.com/BCN3D/BCN3D-Moveo
17. Paz BM (2013) Gamification: a tool to improve sustainability efforts. A dissertation Submit-
ted to the University of Manchester for the Degree of Master of Science in the Faculty of
Engineering and Physical Sciences
18. Ahmad NL, Omhover J (2014) Game methodology for design methods and tools selection. J
Learn Des 7(1):1–9
 An Improved Robot Path Planning Algorithm
for a Novel Self-adapting Intelligent Machine
Tending Robotic System

Feiyu Jia, Juan Tzintzun, and Rafiq Ahmad(B)

University of Alberta, Edmonton, AB T6G 2R3, Canada

rafiq.ahmad@ualberta.ca

Abstract. Autonomous industrial mobile manipulation systems (AIMMS) are

widely used in manufacturing processes. AIMMS can help with part handling and
delivering, part insertion and extraction, loading and unloading, and some other
auxiliary tasks in a machine tending workshop environment. However, nowadays
most AIMMS cannot truly realize the complete automation because the path for
the mobile platform needs setting in advance in terms of different environment and
inspection need to be executed to ensure safety due to lack of proper path planning
algorithms. Therefore, this paper proposes an improved path-planning algorithm
based on Rapidly-exploring Random Tree (RRT) and the quintic B-spline curve
technique to generate a collision-free and smoother path for our designed Novel
Self-adapting Intelligent Machine Tending Robotic System in the workspace. In
the end, the proposed algorithm is demonstrated to generate paths for five different
scenarios to test its performance and reliability.

Keywords: Autonomous industrial mobile manipulation systems · Machine

tending · Path planning · Quintic B-spline curve

1 Introduction

Recently, robots are widely used in industry to perform repetitive, dumb, dangerous, dull,
and dirty tasks. Therefore, industry robotics forms an essential part of the manufacturing
backbone [1]. Task flexibility and robotic mobility are two main advantages that mobile
manipulators bring to industrial applications. Furthermore, a mobile robot shall adapt
to changing environments and perform a variety of tasks. It is essential to combine
locomotion capabilities with manipulation abilities to extend the capabilities of future
robotics. The concept of mobile manipulation refers to a robotic system that integrates
a mobile platform, a robot arm, supported by a vision system. “Mobile manipulation” is
introduced by Schuler in 1984 [2], where a robot arm is mounted on a mobile platform
to execute tool handling and delivering duty in a workshop environment.
This paper focuses exclusively on the machine-tending domain by developing an
improved path-planning algorithm for our designed robotic system. Path planning is
a purely geometric process that is only concerned with finding a collision-free path

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 53–64, 2020.
https://doi.org/10.1007/978-3-030-45402-9_7
54 F. Jia et al.

regardless of the feasibility of the path [3–5]. Early efforts to develop deterministic
path planning techniques showed that it is computationally demanding even for simple
systems [6]. Recently, sampling-based path planning (SBP) algorithms have received
considerable attention due to their capability in complex and time-critical real-world
planning problems [7]. SBP is unique in the fact that planning occurs by sampling the
configuration space (C-space). In a sense, SBP attempt to capture the connectivity of
the C-space by sampling it. This randomized approach has its advantages in terms of
providing fast solutions for severe problems. Furthermore, the computational capacity
of the robot controller unit is limited in my applications; SBP is a better choice because
of its lower computational cost.
Sampling-based planning is by no means a novel concept in robotics [8]. It was pro-
posed to overcome the complexity of deterministic robot planning algorithms for a robot
with six degrees of freedom. The use of random computations to solve otherwise rather
tricky problems have been immensely successful [9, 10]. Both sampling-based planners
and the success of random computations inspired the development of the Randomized
Potential Planner (RPP) [11]. RPP used random walks to escape local minima of the
potential field planner. Later on, a planner based entirely on random walks, with adaptive
parameters, was proposed [12].
The work of Barraquand and Latombe [11] paved the way for a new generation
of path planning algorithms that employ randomization. Arguably, the most commonly
used SBP algorithms are Probabilistic Roadmap Method (PRM) [13–15] and Rapidly-
exploring Random Trees (RRT) [16]. Several other algorithms developed at the same
time that outperformed RPP. The intuitive implementation of both RRT and PRM and
the quality of the solutions lead to their widespread adoption in robotics and many
other fields. Although the idea of connecting points sampled randomly is fundamental
in both approaches, these two methods are different in the manner that they construct a
graph connecting the points [17]. A comprehensive survey of work in SBP is presented
[18]. The PRM algorithm has been recorded to implement well in high-dimensional
state spaces. The PRM is created by curves or straight lines that enable the robot to go
anywhere in its free space. The two well-known methods, named visibility graph (VG)
and Voronoi diagram (VD) have achieved very good results for very different types of
roads. However, the path generated by VG touches obstacles at the vertices or edges and
thus is dangerous for the robot. Contrary, the solution paths based on VD are not optimal
because it tends to maximize the distance between the robot and the obstacles.
RRT has received a considerable amount of attention, because of its computational
efficiency and effectiveness and its ability to find a feasible motion plan relatively quickly,
even in high-dimensional space [19]. In [20], RRT is implemented and combined with
a slightly modified dynamic window method. RRT explores a robot working area by
incrementally building a tree, creating new branches by generating points randomly
and linking them to the closest location for which an obstacle-free path is obtained.
A problem in RRT is that it produces paths with many branches in the workspace by
 An Improved Robot Path Planning Algorithm 55

using the randomized technique. To overcome this problem, an improved robot path-
planning algorithm based on the RRT and quintic B-spline curves is proposed to generate
collision-free and smooth paths for a novel self-adapting intelligent machine tending
robotic system.
The paper is organized as follows: In Sect. 2, a novel self-adapting intelligent machine
tending robotic system is introduced. The principle of the algorithm is demonstrated in
Sect. 3. In Sect. 4, the results and discussion are presented. Finally, Sect. 5 concludes
the paper.

2 Self-adapting Intelligent Machine Tending Robotic System

A mobile manipulation system, named self-adapting intelligent robotic system, is

designed for machine tending in the manufacturing field, as shown in Fig. 1.

Fig. 1. The CAD model of the self-adapting intelligent robotic system

The linear actuator concept is designed to optimize the operation stability for robotic
arm movements. This mobile platform has a wide-body design to prevent any tipping
during the loading operation. By eliminating the need for lifting for arm operation, the
vehicle stabilization is executed by four sliding t-slot frames that are driven by four short
traveling linear actuators.
By using short traveling distance linear actuators, the vehicle can uphold to 225 kg
extra payloads such as batteries, sensors, and laboratory (workshop) tools. Because these
heavy duty linear actuators, the design could easily be expanded for larger vehicle body
and more sophisticated robotic arms. The stabilizing mechanism of the design is shown
in Fig. 2.
56 F. Jia et al.

Fig. 2. Stabilizing mechanism

The mobile platform is designed by using a combination of four Mecanum wheels.

A conventional Mecanum wheel has several rollers that are positioned at an angle of
45° to the central wheel axis [2]. The angled rollers allow the vehicle to translate a part
of rotational force into the normal direction of the wheel. By controlling the rotational
speed of the individual Mecanum wheel, the resultant translated force of four-wheel
combination can determine the direction of the motion for the vehicle without changing
the direction of the wheel [21]. A demonstration for basic four-wheel Mecanum wheel
motions is shown in Fig. 3.

Fig. 3. Basic motions in v-x Axis [21]

By using Mecanum wheels, the vehicle can travel in any direction under any orienta-
tion. This allows the vehicle to break the non-holonomic constraints [2]. The purpose of
this design is to develop an autonomous mobile platform for workshop and laboratory,
which is often an environment with dynamic obstacles and confined spaces. To accom-
plish vehicle movement without complicated maneuvers and path-planning in such a
narrow space, the omnidirectional character of Mecanum wheel has become the best
solution for the design. The bottom view of the mobile platform is shown in Fig. 4.
 An Improved Robot Path Planning Algorithm 57

Fig. 4. Bottom view of mobile platform

In order to control individual motor precisely, two dual-channel controllers are used.
By using a controller area network (CAN bus), one controller will operate as the master,
and the other will serve as the slave [22]. A systematic drawing for the Mecanum wheel
motion control is shown in Fig. 5.

Fig. 5. Systematic illustration of CAN bus [22]

3 Outline of the Proposed Algorithm

3.1 Principle of the RRT Algorithm

To introduce the proposed algorithm, the necessary procedures of RRT must be defined.
They are demonstrated as follows: (1) sampling: this procedure is used to select a config-
uration randomly and add it to the tree. The samples can be either in the free, or obstacle
configuration space. This is the core of the RRT; (2) metric function: this procedure
returns a value, or cost that indicates the effort required to reach C1 from C2, given two
configuration C1 and C2 in workspace; (3) nearest neighbor: this procedure uses a search
algorithm which returns that closest node to the new sample. The value is based on the
predefined metric function; (4) select parent: this procedure selects a nearest existing
node to connect to the newly sampled node. That current node is considered parent; (5)
local planning: given two configurations C1 and C2, this procedure attempts to establish
58 F. Jia et al.

a connection between them; (6) collision checking: it is mostly a Boolean function that
returns success or failure when connecting two configurations C1 and C2. A connection
is successful if it does not intersect an obstacle area [18].
The ability of RRT to explore free space in a workspace with obstacles is illustrated
in Fig. 6. Two Red Cross represents the start and goal point. The green straight-line is
the tree or path generated by the RRT algorithm.

Fig. 6. RRT

It is known that the RRT algorithm is intuitive to employ straight-line paths, and
the tree produced contains too many branches. However, for most robotic systems this
is not a feasible plan due to kinematic or dynamic constraints. Thus, we integrate the
quintic B-spline curves to help the robotic system generate a curvature continuous path
in a practical workspace.

3.2 Path Generation Using Quintic B-spline Curves

In this part, we use a quintic B-spline curve to interpolate points Qk , k = 0…n. The
quintic B-spline curve is presented as follows:


m
S(t) = P i bi,5 (t) (1)
i=0

Where P i is the control point, m = n + 4, bi,5 (t), i = 0, …, m, is a quintic B-spline basis

function. According to the chord length method, if an interpolating curve follows very
closely to the data polygon, the length of the curve segment between two adjacent data
points would be very close to the length of the chord of these two data points. The length
of the interpolating curve would also be very close to the total length of the data polygon.
Then, we assign parameter value τi to each data point Q i . It is shown as follows:

τ0 = 0 (2)
 
 Q i − Q i−1 
τi = τi−1 + , i = 1, . . . , n (3)
L
 An Improved Robot Path Planning Algorithm 59

Where L is the sum of the lengths of these chords:


n
 
L=  Q i − Q i−1  (4)
i=1

Therefore, the ratio of the chord length from the data point Q 0 to data point Q k , denoted
as L k , over the length of the data polygon is
k  
 
i=1 Q i − Q i−1
Lk = (5)
L
For a clamped quintic B-spline curve, we compute a knot vector for interpolation
without end conditions [23]:

t0 = . . . = t5 = 0, tn+1 = . . . = tn+6 (6)

1
j+6
t j+5 = τi , j = 1, . . . , n − 5 (7)
5
i= j

The performance of the quintic B-spline curve to smooth a straight-line path is shown
in Fig. 7.

Fig. 7. Quintic B-spline curve

4 Results and Discussion

In this paper, five different scenarios (the goal point in each scenario in a different
position) are executed to evaluate the developed path-planning algorithm. In each figure,
the black circle areas represent the obstacles in the workspace. The green line represents
the tree, the yellow line is the original path generated by RRT algorithm, the blue line is
trimmed path, and the red is the smooth path produced by our path-planning algorithm.
The results are presented as belows (Figs. 8, 9, 10, 11 and 12):
60 F. Jia et al.

Fig. 8. Scenario 1

Fig. 9. Scenario 2

Fig. 10. Scenario 3

 An Improved Robot Path Planning Algorithm 61

Fig. 11. Scenario 4

Fig. 12. Scenario 5

Though by combining the quantic B-spline curve with RRT can find the shorter
and smoother path, and make it feasible for vehicle motion according to the kinematic
and dynamic constraints, there still exists a chance that the vehicle might collide with
obstacles. Therefore, continuous collision checking must be executed. A rectangular is
placed in the simulation environment as a mobile robotic system, and detection about
the collision between the four edges and the boundary of the obstacles is performed.
When a path cannot be generated, the error message will be reported.
If there is no collision, the path generated. After the path is generated, the robotic
system moves along the path at a constant rate, and its orientation coincides with the
tangent of the curve. The algorithm is applied to three cases to evaluate the performance.
The simulation is shown as follows (Figs. 13, 14 and 15):
62 F. Jia et al.

Fig. 13. Case 1

Fig. 14. Case 2

Fig. 15. Case 3

 An Improved Robot Path Planning Algorithm 63

The results show that the proposed path-planning algorithm has excellent perfor-
mance for the mobile robotic system in a 2-D virtual environment from the above three
cases.

5 Conclusions
In this paper, we propose an improved path-planning algorithm based on quintic B-spline
curve to effectively generate smooth and collision-free paths for our designed Novel Self-
adapting Intelligent Machine Tending Robotic System. The advantage of our algorithm
is that it makes the original path trimmed and smooth, so the mobile robotic system can
continuously move following the kinematic and dynamic rules, and slightly rotate its
wheel, which extends the life of tires. However, this algorithm is not suitable for a very
narrow path between obstacles.
The next stage, our laboratory model will be created in V-REP and the novel
self-adapting intelligent machine tending robotic system be placed in this 3-D virtual
environment to test the performance of the proposed path planning algorithm.

References
1. Hvilshøj M, Bøgh S, Nielsen OS, Madsen O (2012) Autonomous industrial mobile
manipulation (AIMM): past, present and future. Ind Robot Int J 39(2): 120–135
2. Adascalitei F, Doroftei I (2011) Practical applications for mobile robots based on mecanum
wheels - a systematic survey. Technical University of Iasi
3. Wallace R, Stentz A, Thorpe C, Moravec H, Whittaker W, Kanade T (1985) First results in
robot road following. In: Proceedings of the IJCAI, pp 381–387
4. Antonelli G, Chiaverini S, Fusco G (2007) A fuzzy-logic-based approach for mobile robot
path tracking. IEEE Trans Fuzzy Syst 15(2):211–221
5. Jazar RN (2010) Mathematical theory of auto-driver for autonomous vehicles. J Vib Control
16(2):253–279
6. Reif JH (1979) Complexity of the mover’s problem and generalizations. In: Proceedings of
the 20th annual symposium foundation computer science, October 1979, pp 421–427
7. Lee J, Kwon O, Zhang L, Yoon SE (2014) A selective retraction-based RRT planner for
various environments. IEEE Trans Robot 30(4):1002–1011
8. Donald BR (1987) A search algorithm for motion planning with six degrees of freedom. Artif
Intell 31(3):295–353
9. Metropolis N, Ulam S (1949) The Monte Carlo method. J Amer Statist Assoc 44(247):335–
341
10. Metropolis N, Rosenbluth AW, Rosenbluth MN, Teller AH, Teller E (1953) Equation of state
calculations by fast computing machines. J Chem Phys 21(6):1087–1092
11. Barraquand J, Latombe J-C (1991) Robot motion planning: a distributed representation
approach. Int J Robot Res 10(6):628–649
12. Carpin S, Pillonetto G (2005) Motion planning using adaptive random walks. IEEE Trans
Robot 21(1):129–136
13. Kavraki L, Latombe JC (1994) Randomized preprocessing of configuration space for path
planning: articulated robots. In: Proceedings of the IEEE/RSJ/GI international conference on
IROS, vol 3, September 1994, pp 1764–1771
64 F. Jia et al.

14. Amato NM, Wu Y (1996) A randomized roadmap method for path and manipulation planning.
In: Proceedings of the IEEE international conference on robotics and automation, vol 1, April
1996, pp 113–120
15. Kavraki LE, Svestka P, Latombe JC, Overmars MH (1996) Probabilistic roadmaps for path
planning in high-dimensional configuration spaces. IEEE Trans Robot Autom 12(4):566–580
16. LaValle SM (1998) Rapidly-exploring random trees: a new tool for path planning. Dept.
Comput. Sci., Iowa State Univ, Ames, IA, USA, Technical report, TR 98-11
17. Karaman S, Frazzoli E (2011) Sampling-based algorithms for optimal motion planning. Int J
Robot Res (IJRR) 30(7):846–894
18. Elbanhawi M, Simic M (2014) Sampling-based robot motion planning: a review. IEEE Access
2:56–77
19. Hsu D, Latombe JC, Kurniawati H (2006) On the probabilistic foundations of probabilistic
roadmap planning. Int J Robot Res (IJRR) 25(7):627–643
20. Murtra AC, Trulls E, Sandoval O, Perez-Ibarz J, Vasquez D, Mirats-Tur JM, Ferrer M, Sanfeliu
A (2010) Autonomous navigation for urban service mobile robots. In: IEEE/RSJ International
Conference on Intelligent Robots and Systems, IROS, Taipei, Taiwan, pp 4141–4146
21. NEXUS Robot (2016) Mecanum wheel - NEXUS Robot, September 2016
22. RoboteQ (2015) AN1543-Driving Mecanum Wheels, 25 October 2015
23. Piegl L, Tiller W (2000) Curve interpolation with arbitrary end derivatives. Eng Comput
16:73–79
 Stability Analysis of a Rotor Systems
with Flow Forces

Jaime Cruz Cruz1(B) , Miguel Toledo Velázquez1 ,

and Oliver M. Huerta Chávez2
1
SEPI ESIME Zacatenco IPN, 07580 Ciudad de México, Mexico
jcruzc0500@alumno.ipn.mx, mtv49@yahoo.com
2
SEPI ESIME Ticoman IPN, 07580 Ciudad de México, Mexico
lecram 21@hotmail.com
https://www.sepi.esimez.ipn.mx/,
https://www.sepi.esimetic.ipn.mx/

Abstract. The vibration that occurs in Turbomachinery modifies

parameters in its performance and depending on the application, such
performance can compromise the entire system. The vibrations that
occur in these cases can be divided depending on the source that exits, it:
forced and self-excited, where the self-excited vibration extracts energy
from the source, which generates instability and can cause catastrophic
failures since it is very difficult to predict these effects during the design
of the elements of the turbomachine. There are studies to predict some
effects, from energy methods to coupled solutions Fluid-Structure and
by using numerical solutions with neural networks that estimate the
forces to which the rotor system is subjected, so it is It is important
to investigate the effects of these forces on the structural response of the
rotor system [5], therefore fluid-structure interaction studies (such as
those applied in aircraft structures) are needed for rotodynamic insta-
bility. Therefore, this work presents a structural model of a rotor system
considering loads due to the flow to obtain the aerodynamic forces and
analyse the structural stability of the rotor system, numerical and ana-
lytical solutions have been chosen to see the behaviour of the system.
The CFD solution is made to obtain the forces due to the flow in the
rotor system and later these forces feed the structural model of the same
to obtain its response (unbalance) using the finite element method.

Keywords: Stability · Rotor dynamics · CFD · Flow · Unbalance

1 Introduction

The problems associated with vibrations in compressors and turbines have

become more frequent in recent years. This is partly due to the turbomachine’s
tendency towards higher speed and power units that operate well above the first
critical speed. In particular, some recent high pressure centrifugal compressors
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 65–75, 2020.
https://doi.org/10.1007/978-3-030-45402-9_8
66 J. C. Cruz et al.

have had serious non-synchronous vibrations. Recent analyzes on stability and

destabilizing forces associated with couplings, bearings, seals, impellers, etc.,
attest to the importance of the problem. Within the current state of knowl-
edge, a stability analysis of rotor bearing systems can be performed through
the available software. The simplest analyzes are based on the energy dissipa-
tion associated with the dampened free vibrations of the system, while the more
advanced analyzes consider real-time transient solutions in the rotor and can
incorporate non-linear rolling characteristics, destabilization forces induced by
the impeller, non-symmetrical supports, etc [7].
The purpose of this work is to develop a simplified stability analysis for a
flexible rotor in rigid bearings based on the dampened free vibrations of the
system due to the surface forces that are generated due to the interaction with
the flow [6]. The analysis provides a quick method to establish guidelines during
the design of the rotor and to compare existing rotor bearing systems, since the
forces caused by the flow generate stresses whose resultant modifies the critical
speeds when compared with a constant mass analysis. Although this procedure
incorporates several simplifying assumptions and is obviously not intended to
replace advanced analyzes (such as bearing instrumentation and aeroacoustic
analysis), it is useful when considering system damping, a parameter that is an
important factor with respect to stability [3].

A. Assumptions

The following simplifying assumptions are invoked in the present analysis.

– The stability criterion is based on the dampened free vibrations of the rotor
bearing system (with cross coupling rotor forces ignored) at the first critical
frequency.
– They are considered ideal bearings. (The effects of cross-coupling of the bear-
ing are neglected depending on the inherent stability of the bearing of the tilt
pad with respect to rotation).
– The rotor is supposed to be symmetric.
– In addition, vibration at the joints of the supports with the shaft is not
considered.

2 Spring-Mass Model
The equations of movement and response of the rotor for the synchronous move-
ment of an imbalanced Jeffcott rotor (massless axis, mass concentrated in half of
the space) in linear isotropic bearings and flexible supports have been derived.
In particular, it is shown that the response with rigid pedestals has an identical
shape to that of a simple and one-dimensional spring-mass system [5].

F
= KE + iωC − MS ω 2 (1)
X1
 Stability Analysis of a Rotor Systems with Flow Forces 67

Fig. 1. Spring mass model

Fig. 2. Spring mass model 2D

Where    2
K
K2 1+ K2 + Kω (C22 K1 +C12 K2 )
KE =  1
K2
2  1 2 (2)
1+ K + Kω (C1 +C2 )2
1 1
   2
K2
C1 +C2 + Kω
K1 C1 (C22 +C12 C2 )
CE =   
K2 2
1
ω
2 (3)
1+ K + K (C1 +C2 )2
1 1

Since Eq. (1) is identical in shape to the analog equation for the simple spring-
mass damping model shown in Fig. 2, then the spring and the damper associated
with a simple equivalent model of the two-level system given by Eqs. (2) and (3),
[4] develops the response of the simple system in Fig. 2 to a rotating unbalance
of (Mse) as
MS eω 2
X= (4)
(KE − MS ω 2 )2 + (CE ω)2
Equations (2), (3) and (4) are identical to the dynamic results of the rotor
with zero-axis damping, that is, with C \ = 0. Considering that the simple spring
68 J. C. Cruz et al.

mass damping system of Fig. 2 is an equivalent model of a rotor bearing system,

an important parameter is the classical damping factor f, defined as
dynamic resonant displacement 1
Q=  (5)
zero frequency displacement 2ζ
As indicated above, the slightly dampened spring mass damping system
(f < 0.2) shown in Fig. 2 will experience free vibrations or forced vibrations res-
onant at almost its natural frequency (not dampened), that is

KE
Response Frequency  ωn = (6)
MS

Since Ke is a frequency function, the determination of w, in Eq. (7) implies

a fourth order expression. When solving the frequency, the parameters KE, CE
and f can be evaluated from Eqs. (2), (3) and (5), respectively.
However, as the damping of the system decreases towards zero, the param-
eters KB, CE and, therefore, f become independent of the frequency and the
analysis is considerably simplified. Since it has been stated that stability can
become a factor of f < 0.055, it is appropriate to investigate a first order distur-
bance on this simple type solution to establish the application limits [3].
Considering a rotor bearing system with zero-axis damping (Ci = 0) and
using Eqs. (5) and (7), the expressions for KE and CE given by Eqs. (2) and (3)
become [1]   
K K C 2
1+ 1+ K2 γ CE 2
K1 K2
KE = K1 +K2 = 1 E
KE C2 2
(7)
1+ γ C
E
 
C2 
CE = 
K 2  1 
KE C2 2
(8)
1+ K2 1+ γ C
1 E

where
γ= 2ζ
(K1 +K2 ) (9)
As the damping factor tends to very small values, Eqs. (8) and (9) become
K1 K2
KE0 ≡ KE | ζ → 0 = K1 +K2 (10)
C2
CE0 ≡ CE | ζ → 0 = (1+K2 /K1 )2 (11)
Substituting the above “zero” order solutions into the brackets of (8) and (9)
results in the first order corrected solutions
A
KE1 = KE0 B (12)
C 
CE1 = B
E0
(13)
Where  
2
K2 K1
A=1+4 K1 (ζ0 )2 1 + K2
(14)
 Stability Analysis of a Rotor Systems with Flow Forces 69

 2
K2
B =1+4 K1 (ζ0 )2 (15)

ζ0 = √ CE0 ≡ CE0
(16)
2 MS KE0 2MS ωn0

From the given expressions, the ratios between the first and zero order
approximations for ωn and ζ are

ωn1 A
= (17)
ωn0 B
ζ1 1
=√ (18)
ζ0 AB
Thus, first order corrections to the “zero” order solution (which assumes very
light dumping) may be obtained through Eqs. (18) and (19) if required.

3 Fluid Forces
The loads due to the flow that are presented in a rotating element depend on the
angular velocity to which it is subjected since velocity components are generated
that modify [2] the generated forces and the flow fields have a different behavior
for each velocity value (Fig. 3).

Fig. 3. Blade fluid

For a three-dimensional model a rotor model like the one shown in the figure
is used, with the obtained equations, the geometry of a rotor for finite element
analysis and CFD respectively, where the initial model is as shown (Fig. 4):
70 J. C. Cruz et al.

Fig. 4. Rotor model

With a discretization in mixed elements you get a mesh for structure and
CFD analysis as shown in the figure (Fig. 5).

Fig. 5. Rotor mesh

 Stability Analysis of a Rotor Systems with Flow Forces 71

Using the same configuration as in Fig. 1, and defining ideal supports and
an angular speed of 3600 rpm, a behavior of forces was obtained as shown in
the figure displacement in the mode 2. Where the displacements due to the flow
produces a lateral displacement in the axis, which in turn modifies the flow field
in the rotor (Figs. 6, 7, 8 and 9).

Fig. 6. Streamlines

Fig. 7. Deformation
72 J. C. Cruz et al.

Fig. 8. Displacement x

Fig. 9. Displacement y

4 Stability Criterion

Having established a means of determining the damping factor as a function of

the rotor-bearing system parameters, it is appropnate to consider guidelines as
to the values of the damping factor which can be associated with rotor stability.
For this consideration, the log decrement “o” is a useful parameter. For a simple
mass-spring-damper system (Fig. 2) with light damping

δ  2πζ (19)
 Stability Analysis of a Rotor Systems with Flow Forces 73

Un − Un+1
= 1 + e−2δ (20)
Un
U R1 cycle = (U − U ) + U e−2δ (21)
where U n + 1 is the energy associated with the (n + l)th cycle. Assume now that
a constant amount of external energy “U” is supplied to the rotor-bearing system
during each cycle of v1bration and that the dissipation function opera’tes on the
total energy existing at the end of the previous cycle but only on a P?1·tion “E”
of the energy supplied during the particular cycle. ’l hus the energy remaining
in the system after one cycle is

U Rncycles = U 1 + e−2δ + e−4δ ...........e−2(n−1)δ − U (1 − e−2nδ ) (22)

and after cycles

U Rncycles = U 1 + e−2δ + e−4δ ...........e−2(n−1)δ − U (1 − e−2nδ ) (23)

in the limit, as the number of elapsed cycles become very large (i. e. n → ∞)

1
U Rn→∞ = U − (24)
1 − e−2δ

In order to satisfy the boundary condition of U Rn→∞,δ→∞ = 0, the param-

eter  must equal on (1.0), and hence

e−2δ
U Rn→∞ = U −  , where δ = 2πζ (25)
1 − e−2δ

Considering that detrimental nonsynchronous vibrations occur due to an

accumulation of energy by the system as opposed to the specific quantity of
input energy U associated with each cycle. For instance, if the destabilizing
forces are small (small energy input per cycle). The usefulness or need of a more
advanced analysis where required.

5 Sample Calculations

Sample calculations for a high-speed compressor rotors are shown using the loads
due to the flow obtained in the CFD analysis. A numerical approximation was
made and the behavior shown in the figures was obtained (Figs. 10 and 11).
74 J. C. Cruz et al.

Fig. 10. Unbalance by fluid force

Fig. 11. Unbalance by fluid force

6 Conclusions
The paper presents a procedure for the analysis of stability due to forces gener-
ated by the flow in turbomachinery elements. The pressure excitations calculated
by the harmonic methods of CFD are assigned to the structural model where
a frequency response analysis is performed. This way, the influence of pressure
fluctuations is estimated in the flow in the rotor towards the lateral displace-
ments in the axis. In a post-processing stage an evaluation of for life that takes
into account the vibrations induced by the flow. The procedure is considered for
the use of models newspapers that are often applied in simulations of turboma-
chinery The vibration responses in the blades to the pressure fluctuations are
calculated and the points are located of greater deformation in the axis. The
procedure presented can be easily transferred to other areas of application, such
 Stability Analysis of a Rotor Systems with Flow Forces 75

as electromagnetically induced vibrations in motors or generators. A result of

the frequency response analyzes can also be the sound pressure level for acoustic
evaluations.

References
1. Genta G (2005) Dynamics of rotating systems. Mechanical engineering series.
Springer, New York
2. Friswell MI, Penny JET, Lees AW, Garvey SD (2010) Dynamics of rotating
machines. Cambridge University Press, Cambridge
3. Rao SS (2016) Mechanical vibrations. Prentice Hall, Upper Saddle River
4. Dimarogonas AD, Paipetis SA, Chondros TG (2013) Analytical methods in rotor
dynamics. Springer, New York
5. Krämer E (2013) Dynamics of rotors and foundations. Springer, Heidelberg
6. Khader N (1997) Stability analysis for the dynamic design of rotors. J Sound Vib
207(3):287–299
7. Sinha SK (2004) Dynamic characteristics of a flexible bladed-rotor with Coulomb
damping due to tip-rub. J Sound Vib 273(4–5):875–919
 Flutter Analysis of Rotor Based
on a Fluid–Structure Method

Jaime Cruz Cruz2(B) , Miguel Toledo Velázquez1 , Oliver M. Huerta Chávez2 ,

Gibran Jalil Garnica Castro2 , and Rafael Sánchez López1
1
SEPI ESIME Zacatenco IPN, 07580 Ciudad de México, Mexico
mtv49@yahoo.com
2
Tecnológico de Estudios Superiores de Ecatepec, 55210 Estado de México, Mexico
jcruzc0500@alumno.lpn.mx, lecram 21@hotmail.com
http://www.ipn.mx, http://www.tese.edu.mx

Abstract. The vibration that occurs in every rotating element (specifi-

cally in turbomachinery elements) generates effects that modify its oper-
ation and therefore its performance, which causes the system to compro-
mise and not operate in nominal conditions. Of all the types of vibra-
tion that can occur due to various sources of excitement: body or flow
scattering, what is clear is the generation of a forced vibration: where
a self-excited vibration is generated and whose source energy It gener-
ates dynamic instability, which can lead to catastrophic failures due to
the complexity of predicting these effects during design. Recent studies
focus on aeroelasticity in the rotating elements, when using this app-
roach, the elasticity equations and the aerodynamic forces generated in
the blades are used, thus obtaining the equations of aeroelasticity, thus
these equations are discretized and it is solved numerically to obtain the
corresponding approximations.
The present work presents a structural model of a rotor considering
the shifts that are presented by the fluid, for this case the complete
rotor is studied not before analyzing the behavior of the flow in a blade,
thereby obtaining displacements in the structure due to the flow and
with it the vibration modes in the blade. The analysis in the blade is
performed by FEM and CFD to obtain the behavior of the vibration
modes considering the flow.

Keywords: Fluid-structure · Rotor dynamics · CFD · FEM

1 Introduction
The tendency of increasing the efficiency of gas-turbine engines implies the neces-
sity of designing wide-chord unshrouded fan blades of complex spatial aero-
dynamic shape. However, with such blades, the risk of aeroelastic instability
(e.g., flutter) development increases. For reliable and safe operation of gas-
turbine engines with such fans, it is necessary to be able to accurately predict
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 76–84, 2020.
https://doi.org/10.1007/978-3-030-45402-9_9
 Flutter Analysis of Rotor Based on a Fluid–Structure Method 77

the aeroelastic behavior of fan blades at the design stage. The conventional app-
roach to the utter calculations of the bladed disks is based on the frequency
analysis method assuming that blade motions are described by harmonic time-
dependent functions with a constant phase shift between the adjacent blades,
whereas the natural modes and frequencies of the rotor low system remain sim-
ilar to those in vacuum. This method ignores the feedback effect of the gas ow
on blade oscillations.
This paper presents the solution methodology of the unsteady multidimen-
sional aerodynamic problem coupled with the elastic blade oscillation problem
in the gas ow. Also presented are the results of calculations for aeroelastic char-
acteristics of a fan bladerow (BR) of a modern bypass gas-turbine engine.

2 Governing Equations

On the one hand, when considering the gas ow BR with oscillating blades, one has
to take into account that the gasdynamic parameters depend on the blade veloc-
ity and position determined by a specied oscillation law (kinematically forced
oscillations). On the other hand, when considering the problem of forced oscilla-
tions, the blade velocity and positions depend on the aerodynamic loads forcing
the blades.
The adequate accounting of energy exchange can be attained by solving a
coupled aerodynamic flow aeroelastic oscillations problem, based on simulta-
neous integration, using the time-marching scheme, of the complete system of
equations of unsteady aerodynamics and aeroelastic blade oscillations [3,5].
The three-dimensional transonic ow of inviscid gas with zero thermal con-
ductivity through the axial turbomachine is considered in the physical domain,
including a fan BR, rotating at a constant angular velocity. The ow is described
by the complete system of unsteady Euler equations presented in the integral
form [4].   
∂ →→
−
f dΩ + F − n dσ + HdΩ (1)
∂t
Ω σ Ω

where ⎡ ⎤ ⎡ ⎤
ρ ρ−
→
υ
⎢ ρυ1 ⎥ ⎢ ρυ1 −
→
υ + δ1i p ⎥
⎢ ⎥ → ⎢
− →
− ⎥
f =⎢ ⎥
⎢ ρυ2 ⎥ F =⎢ ρυ
⎢ 2− υ + δ2i p ⎥
⎥
⎣ ρυ3 ⎦ ⎣ ρυ3 →
υ + δ3i p ⎦
E (E + p)−
→υ

⎡ ⎤
0
⎢ ρae1 − 2ρωυ2 ⎥
⎢ ⎥ 1, j=i υ12 + υ22 + υ32 + r2 ω 2
H=⎢ ⎥
⎢ ρae2 + 2ρωψ2 ⎥ δji = E = ρ(ε + )
⎣ 0 ⎦ 0, j = i 2
0
78 J. C. Cruz et al.

Ω is the volume of the finite-volume computational cell with moving bound-

aries σ; −
→n is the unit exterior normal; p and ρ are the pressure and density,
respectively; ω is the angular rotation velocity; υ1 , υ2 and υ3 are the compo-
nents of the velocity vector −→υ ; ae1 and ae2 are the projections of traslational
acceleration; E is the total energy per unit volume; ε is the internal energy per
unit mass; and r is the distance to the rotation axis.
The system of Eqs. (1) is supplemented with the ideal gas equation of state:
1 p
ε=
k−1ρ
where k is the specific heat ratio.
Equations (1) are integrated in the computational domain, including the
complete circumference, i.e., the number of interblade channels is equal to the
number of blades. The number of interblade channels, N , and interblade phase
angle δ are related to each other as N δ = 2πj (where j is the integer number). [2].
The computational grid is divided into z segments, either of which includes
one blade and possesses the circumferential length equal to a rotor pitch. Each
segment is discretized using the hybrid H − H grid for rotor channels. The
external H-grid remains fixed, while the internal H-grid is reconstructed at every
iteration in accordance with the position of the oscillating blade [7]
The discrete form of Eqs. (1), obtained for an arbitrary spatially deforming
grid [1]:
1
Δt f i+1/2,j+1/2,k+1/2 Ω i+1/2,j+1/2,k+1/2 − fi+1/2,j+1/2,k+1/2 Ωi+1/2,j+1/2,k+1/2
+ − (f σωn )i+1 + (f σωn )i − (f σωn )j+1 + (f σωn )j − (f σωn )k+1 + (f σωn )k
+ (F1 σ)i+1 − (F1 σ)i + (F2 σ)j+1 − (F2 σ)i + (F3 σ)k+1 − (F3 σ)k
+Hi+1/2,j+1/2,k+1/2 Ωi+1/2,j+1/2,k+1/2 = 0

Where subscripts and superscripts stand for “new” and “old” cells; f =
{ρ, ρ−
→
υ , E} is the symbolic vector of unknown variables; F1 , F2 and F3 are the
values of variables in the centers of cell faces; σ and ωn are the face area and
normal velocity in the face center, respectively. The gasdynamic parameters on
the faces are determined from the solution of Riemann problem with the use of
iterative process [6].
The dynamic model of the oscillating blade in linear formulation is described
by the matrix equation [8]:

[M ] {ü(x, t)} + [C] {u̇(x, t)} + [K] {u(x, t)} = 0

Where [M ], [C], and [K] are mass, damping, and stiffness matrices, respec-
tively; {u(x, t)} is the blade displacement; and [F ] is the vector of unsteady
aerodynamic loads. Using the modal approach:
[ 
{u(x, t)} = [U (x)] {q(t)} =i= 1]N {Ui (x)} qi (t)
 Flutter Analysis of Rotor Based on a Fluid–Structure Method 79

Where Ui (x) is the blade displacement vector in the ith mode, and qi (t) is the
modal coefficient. Based on the condition of orthogonality of natural modes, the
Eq. (2) can be transformed to the system of independent differential equations
with respect to the modal coefficient of natural modes:
q¨i (t) + 2hi q˙i (t) + ωi2 qi (t) = λi (t)
With hi is the damping coefficient of the ith mode; ωi is the natural frecuency
of the ith mode; λi is the modal force corresponding to the displacement within
These ith mode, calculated in the iteration based on the pressure distribution
on the blade. 
pU i ndσ
σ
λi =  2
ρU i dυ
υ
Where ρ is the pressure on the blade surface.

3 Numerical Modeling
The numerical simulation of aeroelastic characteristics was performed for a fan,
consisting of 24 blades (Fig. 1) for the regime with n = 0:7 at the operation
line. The rst ve natural modes and frequencies of blade oscillation are shown in
Fig. 1 Fan bladerow Table 1 (all natural frequencies are related to the rst natural
frequency). The calculations were performed using 10 25 58 (radial tangential
axial) computational grid for each interblade channel.

Fig. 1. Blade rotor CAD

At the first stage, the aerodynamic calculations for a rotating fan BR with
a preset blade oscillation law were made. All blades were assumed to exhibit
harmonic oscillations according to each of the natural modes following the same
law with a constant interblade phase angle (IBPA) Æ:
qij = qi0 sin [2πνi t + (j − 1)δ]
80 J. C. Cruz et al.

where qij is the modal coefficient; i is the natural mode number; j is the blade
number; qi0 is the oscillation amplitude of the ith mode; and i is the natural
frequency.
The mesh created with tetrahedral elements adjusting to the volume of the
blade (Fig. 2).

Fig. 2. Blade rotor Mesh

The calculation of the aerodynamic forces that occur in the wing in the
influence due to the flow generate the oscillations in them and at the same time
generate oscillation frequencies in the blades.
The calculations were performed for harmonic oscillations of blades on a
3600 rpm with the six modes (Fig. 3).

Fig. 3. Pressure distribution

 Flutter Analysis of Rotor Based on a Fluid–Structure Method 81

Of the previous figure, the behavior of the pressure distribution is observed

and the following figure, a graph of its behavior (Fig. 4).

Fig. 4. Pressure

A cascade configuration shows the behavior of a blade to obtain its vibration

modes due to flow loads. For a speed of 3600 rpm the distribution of pressures
in the blade in Fig. 5 is observed.

Fig. 5. Pressure distribution

82 J. C. Cruz et al.

Depending on the angular velocity, the variations change in each mode so

that a whole system must be analyzed in order to obtain its natural frequencies
and thus have data that help the prediction of failures in the rotating elements
(Figs. 6, 7, 8 and 9).

Fig. 6. Six modes of rotor

Fig. 7. Displacement mode 1 and 2

 Flutter Analysis of Rotor Based on a Fluid–Structure Method 83

Fig. 8. Displacement mode 3 and 4

Fig. 9. Displacement mode 5 and 6

4 Conclusions
The present work presented the numerical simulation of the aerolastic character-
istics of the Arabic using the Fluid structure model where flow properties were
found that affect the modal behavior for each frequency.
It was found that there is a damping in the case of harmonic oscillations of
the blades, for each of the natural modes with the interaction between the modes
taken into account, that is, the energy of the oscillating blade is transferred to
the main flow and vice versa as well as a damping due to the flow that can
reduce the amplitudes and oscillation frequencies in the blade by means of the
constant feeding of fluid during the operation of the rotor, that is to say in the
gas-rotor coupling.

References
1. Hughes TJR (2012) The finite element method: linear static and dynamic finite
element analysis. Dover Civil and Mechanical Engineering
2. Genta G (2005) Dynamics of rotating systems. Mechanical engineering series.
Springer, New York
84 J. C. Cruz et al.

3. Blake WK (2017) Mechanics of flow-induced sound and vibration, general concepts

and elementary sources, vol 1, 2nd edn. Academic Press, Cambridge
4. Friswell MI, Penny JET, Lees AW, Garvey SD (2010) Dynamics of rotating
machines. Cambridge aerospace series. Cambridge University Press, Cambridge
5. Paidoussis MP (1998) Fluid-structure interactions: slender structures and axial flow.
Elsevier Science, Oxford
6. Chakrabarti SK (2005) Numerical models in fluid-structure interaction. WIT Press,
Southampton
7. Taylor NV, Allen CB, Gaitonde A, Jones DP (2004) A structure-coupled CFD
method for time-marching flutter analysis. Aeronaut J 108:389–401
8. Sayma AI, Vahdati M, Imregun M (2000) An integrated nonlinar approach for turbo-
machinery forced response prediction part 1: formulation. J Fluids Struct 14:87–101
 Amphibious Propeller Optimization
for Aerial and Underwater Operation

C. Genaro Carreon1(B) , C. Oliver M. Huerta2 , M. Alfredo Arias2 ,

C. Sergio Torres2 , and José Ángel L. Ortega H.1
1 IPN-ESIME Unidad Zacatenco, SEPI, Av. Luis Enrique Erro S/N, Nueva
Industrial Vallejo, 07738 Gustavo A. Madero, Ciudad de México, Mexico
gcarreonc1000@alumno.ipn.mx
2 IPN-ESIME Unidad Ticoman, SEPI, Av. Ticoman 600, San José Ticoman,
07340 Gustavo A. Madero, Ciudad de México, Mexico
lecram_21@hotmail.com, aarias@ipn.mx, sgtorres.c@gmail.com,
https://www.ipn.mx/

Abstract. In this paper a novel methodology for biphasic flow (air-

water) is presented. It involves a CST-parametrization and the optimiza-
tion by NSGA-II coupled to design and optimize amphibious propellers,
through CL, CD and CP coefficients, which allow to operate in aerial and
underwater environment. Finally, the results of CL, CD and CP, were
evaluated by XFoil CFD software before and after of the optimization
and then compared to each other, Evolving the optimization process for
140 generations, give a total of 70000 individuals and obtain a relative
improvement for the new propeller design as nearly as 30%.

Keywords: Propeller · Multiobjetive optimization · CFD simulation ·

Genetic algorithm

1 Introducción

Unmanned vehicles are more present in daily activities. We can find drone
projects from that can deliver mail packages to even the most sophisticated
ones as used in unfriendly zones like hurricanes. However, always we need new
systems that they are able to operate in two or more environment to maximize
their respective operations, minimize response time and reduce production costs
for each vehicle to specific condition.
Proof of it, are the investment did by [9], who made a functional diagram of
different vehicles sittings, whit capabilities to work in two distinct media. The
Fig. 1 shows Drones for two o more media. This paper allowed us to say that
the actual hybrid vehicles are not only concepts, but they already exist and are
developed for various applications.
Supported by LASIRS.
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 85–96, 2020.
https://doi.org/10.1007/978-3-030-45402-9_10
86 C. Genaro Carreon et al.

Fig. 1. Unmanned autonomous vehicle classification for operations in two media [9]

Research like this one has 4 principal aspects to develop: “Structure”, “Elec-
tronics”, “Control” and “Propulsion”. In the propulsion section, this will adapt to
the operation parameters in both media, specially to try to balance and if pos-
sible, maximize the operational performance. In our case, we set an operation
rate as 50% aerial and 50% underwater.
The first problem we find is the different density between air and water
which is nearly 1000 times different. This entails that the propulsion system
has different operation objectives. While the aerial propeller tends to have high
revolutions, the underwater propeller is quite the opposite.
The second problem is, consequence by the first, the cavitation effect of
high revolutions of the propeller, phenomena that reduce the lifecycle and his
performance in underwater operation. In the other hand, in the aerial operation
phase, the working losses are present by compressible effects in the propeller
blade tips.
The third and last problem lies in the optimization itself, since, by improving
the underwater operation, aerial performance is decreased, therefore is necessary
to find the best operation ratio, that would get us the maximum functioning for
both media. To do this, is necessary to make use of a multi-objective optimization
method.
The multi-objective optimization methods allow to find the best values for
two or more variables. Following this idea, we use NSGA-II technique [5]. This
code, which is a NSGA successor, present some advantages:

– Introduction of a non-dominance control.

– Relatively simple setting to adapting our problem.
– Scanning of the most solutions quantities possible.
 Amphibious Propeller Optimization for Aerial and Underwater Operation 87

To realize the optimizations we need a software with abilities to solve the coef-
ficients with the lower time possible to have potentials to analyze the highest
individuals number enough to ensure the population variety during the opti-
mization.
The solution is to use XFoil [6]. XFoil software is based on a panel method for
CFD, it is able to evaluate in 2–4 s, depending of variable number (i.e. Reynolds
number, angle of attack, viscid or inviscid flow, etc). However, the precision that
offer this package is suitable for the study to be realized.
The selected coefficients for this paper are related to lift, drag and pres-
sure. The first two are applied to the aerial operation while the last is related
to the cavitation phenomena through the Thoma coefficient (cavitation coeffi-
cient). This dimensionless value allows us to predict the existence of cavitation
phenomena and its calculation is relatively simple, thus, is the perfect way to
solve the problem mentioned before.

2 Theoretical Background

In this section, we approach the necessary process to set the geometric

parametrization of the airfoil. The objective functions for the optimization and
its relationship with de above point and finally de CFD analysis with XFoil to
calculate the main coeffcients.
There are several types of techniques to use for the parametrization of the
airfoil, which are listed below:

– Bsplines
– Parsec
– Bezier Curves

These are some of the most used procedures for the airfoil’s parametriza-
tion process. However, these approaches have different troubles that make them
nonfunctional for this paper. For example, Parsec [1,7] use airfoils characteristics
having 12 equations with 12 control variables. Nevertheless, this produce a small
design space, additianlly to the complexity to code the equations that define this
method.
In other hand, the Kulfan CST method [2–4] use two functions to represent
many different curves and volumes in 2D or 3D that are used most commonly
in the aircrafts.
ζ(Ψ) = C(Ψ)S(Ψ) + Ψζte (1)
The Eq. 1 is composed of one form and one class function. The form function
indicate what kind of geometry is used; the class function delimits the curve in
the plane (i,j). The strength of this method lies in the Bernstein polynomial,
which one can be worked from second order (8 control variables by airfoil) to
infinite order number.
Nevertheless, the higher the degree of the equation, the greater number
of unknown quantities control and this turn more complex the optimization
88 C. Genaro Carreon et al.

Fig. 2. Bernstein polynomial decomposition used in shape function 1 [4]

method, moreover, increasing the calculation time. For that, a suitable polyno-
mial of grade 4 is used. This gives us 12 variables and obtain a great accuracy
for the geometry parametrization (Fig. 2).
Applying the Bernstein fourth grade polynomial to the form function and
combining in the CTS general equation we obtain the final version of the equa-
tions for the optimization, in which we calculate the Bernstein coefficients to find
the expression that handle the airfoil used in the amphibious propeller. To Find
the coefficients of the Bernstein polynomial fast and suitable, we use a stochastic
search method based in the geometry comparation.
This search method is “fminsearch” MATLAB funcion, and the comparative
geometry function is as follows.
x=i

= abs(zpar − zref ) (2)
x=0

where zpar refers to the distance of the geometry by the parametrization method
while zref indicates the original distance by the initial geometry. This function
is applied both in extrados and intrados airfoil’s surfaces.
For the selection of the optimization we apply the NSGA-II algorithm. This
technique, in GPL license and code, operate under a non-dominated control to
evaluate all individual generated. The use of this methodology is based on the
work done by [8] where they shows all benefits of an analysis of a large number
of individuals, which guarantees a solutions variety and thus be able to conclude
in a optimal solution for the objective functions previously set.
It necessary keep in mind that the optimization techniques always result in
a minimal value, but if the function is coded in specific form, we can obtain as a
result a maximum value. Preserving this in mind, the objective functions must
fulfill the following:

– CL and CP must be a maximum

– CD must be a minimum
 Amphibious Propeller Optimization for Aerial and Underwater Operation 89

The aerodynamical coefficients evaluations must be done in the shortest time

possible in order to analyze as many geometries as possible and, to achieve this,
the XFoil [6] software is used. This software allows to have results oh the right
precision to this problem with the advantage of being fast (in approximately-
break 2–5 s). Another important advantage is its ease of integration with other
softwares like MatLab.

3 Implementation and Results

The mathematical expression that represents the airfoil geometry, is a equation
combining both CST methodology and Bernstein polynomial. This combination
result as next shape function (3).

S(x) = A1 (1 − x)5 + 5A2 (1 − x)4 x + 10A1 (1 − x)3 x2

+ 10A4 (1 − x)2 x3 + 5A5 (1 − x)x4 + A6 x5 (3)

The class function 4, with the coefficients to NACA airfoil by Kulfan proce-
dure are as follows:
C(Ψ) = (Ψ)0.5 [1 − Ψ]1.0 (4)
The Ψ value is equal to x/c where c represent the airfoil chord. To simplify,
the class function is set a unitary chord remaining all the equation depending
only of x.
Combining the Eqs. 3 and 4 into the Eq. 1 is obtained the function value to
represent the position in the “y” axis; while the x values belongs to are taking
by the x positional variable.
This process is inserted in the search method with fminsearch to establish
the value of Ai coefficients since they are the ones that determine the behavior
profile. Initially these quantities are assigned a unitary value, nevertheless, the
ζ value is related to trailing edge and is set to 0.001.
To validate that coefficients found in this step, the Eq. 2 is used, where Zpar
is the value obtained by the parametrization technique in every coordinate x and
is compared to the original position of the airfoil (Fig. 3).

Fig. 3. Graphical representation of function 2 operation

90 C. Genaro Carreon et al.

The execution time of search require nearly 1–3 min taking into account
the maximum iterations number which enables MatLab, the maximum vec-
tor size and the computational resource. The final equation to define the CST
parametrization method is as follows:
 
S(x) = (x)0.5 [1 − x]1.0 A1 (1 − x)5 + 5A2 (1 − x)4 x + 10A1 (1 − x)3 x2

+ 10A4 (1 − x)2 x3 + 5A5 (1 − x)x4 + A6 x5 + xζ (5)

In the other hand, the objective functions are set as follows so that they can
satisfy the goals of the above section.

f (1) = 1 − CL (6)
f (2) = CD (7)
f (3) = 1 + CP (8)

The Thoma coefficient (Eq. 9), dimensionless quantity, link local absolute
pressure, steam pressure and the kinetic energy by volume. This amount
(expressed in negative values) allows to predict the existence of the cavitation
phenomena in the fluid.
p0 − p v PAT M + ρgh − pv
σ= = (9)
0.5ρW 2 0.5ρW 2

The cavitation effect occurs when the lower negative pressure coefficient (CP )
is greater that the Thoma coefficient, for example

−Cp > thoma

−32 > −20

From a strict view point, the CP amount lower that Thoma coefficient ensures
that phenomena cavitation will be present. Therefore, the objective function
related to CP is maximize his rate to avoid cavitation.
The coefficient evaluation was made through XFoil software. This code works
mainly with the Reynolds number, this term helps the operations because only
is needed such number in each designed station to improve. For this paper is
selected 10 stations along to the propeller span to get a quasi-three-dimensional
optimization. Nevertheless, the tip and root stations are omitted because those
give a negligible propulsion work.
For the above reason, only 8 stations remain to optimize of the which are
the conditions of the Table 1. All topics above were implemented in the code
NSGA-II through MatLab language.
The results to the parametrization as shown in Fig. 4. The solution way is
related with the follow flowchart.
Parametrization results are showed in Fig. 5:
Amphibious Propeller Optimization for Aerial and Underwater Operation 91

Table 1. Reynolds number table for the different stations chosen

Station Reynolds number

20% 50620.3900
30% 58952.7626
40% 64505.6269
50% 66193.2478
60% 66301.9375
70% 62530.9932
80% 56496.3187
90% 44574.3631

Fig. 4. Implementation flowchart of the used methodology

Fig. 5. Generated geometries in the parametrization

92 C. Genaro Carreon et al.

For the other hand, analyzing before and after the parametrization all remain
stations is presented in the Table 2.

Table 2. Initial CL values

Station CL
20% 1.052
30% 0.962
40% 1.394
50% 1.330
60% 1.351
70% 1.354
80% 1.204
90% 0.824

The Fig. 6, shows a Pareto Front, which is a graph where are compared the
solutions in which that the optimal is the solution 1 and cannot exits other
solution that improve an objective and worsen other.

Fig. 6. Pareto’s front example

To get a general point of view about the optimization is chosen randomly

some individuals from the outer and the middle of the Pareto front curve to
define how in the far zones only and objective is improved meanwhile in the
intermediate zone is obtained a balance between the objective functions.
The objective functions are next

f (1) = 2 − CL (10)
f (2) = CD (11)
f (3) > −24.72 (12)
 Amphibious Propeller Optimization for Aerial and Underwater Operation 93

To help the optimization and analysis through Pareto’s front, the problem
is sectioned in two parts. The first section is about to maximize the lift to drag
ratio; the second part is increasing the CP value in the selected sections along the
propeller span. The sectioned problem helps to analyze the created geometries
as well as make easier the graph visualization about the objectives and how was
the improvement in coefficients on each environment media work.
The results related to aerodynamical objective, are depicted in the Fig. 7

Fig. 7. Pareto’s front of station 7.

Some individuals selected from the Pareto’s front are given in the Fig. 8
From those airfoils, now is executed the improvement to the pressure coeffi-
cient, substituting the Reynolds number associated to the underwater operation.
From this new calculation, its obtained the Table 3 which specify the Thoma
quantities for each designated station.

Table 3. Table of Thoma coefficients for each station

Station CP
20% −21.0528
30% −19.1875
40% −18.3791
50% −17.3411
60% −15.4947
70% −13.6608
80% −11.8019
90% −11.3029
94 C. Genaro Carreon et al.

(a) Airfoil in left zone (CD y CL (b) Airfoil in right zone (CD y CL
menores) mayores

(c) Airfoil in middle zone (Equilibrio

entre CD y CL )

Fig. 8. Airfoil improved for station 7

Thus, the new airfoils made with this improvement are showed in the Fig. 9
which are the final geometries.
Finally, the comparation between CP values are showed in the Table 4, before
and after geometry optimization.

Table 4. Comparison of pressure coefficient before and after optimization

Station CP before CP after

20% −9.392 −6.1
30% −8.395 −5.35
40% −7.396 −5.026
50% −6.401 −4.61
60% −5.405 −3.87
70% −4.4 −3.135
80% −3.411 −2.39
90% −2.415 −2.19
 Amphibious Propeller Optimization for Aerial and Underwater Operation 95

(a) Airfoil in left zone (CD y CL (b) Airfoil in right zone (CD y CL
menores) mayores

(c) Airfoil in middle zone

(Equilibrio entre CD y CL )

Fig. 9. Airfoil hydrodynamic improved for station 7

4 Conclusions
When doing this work it was possible to set up a new novel methodology for opti-
mization in biphasic fluid to UAV propellers with the 50-50 ratio work objective.
Also, can be proved how, depend which individual for Pareto’s front is selected,
one of the all objective functions is improved. The best option is located in
the middle zone, due to the setting the different objectives. In other way, the
objective functions must be set carefully to made a right selection in the Pareto’s
front. Another important aspect to the optimization is found on the CFD solver.
A code faster with the right accuracy to the problem, help to have a biggest vari-
ety of geometries, ensuring a bigger search space with a lower calculation time.
The Table 4, shows the aerodynamical and hydrodynamical improvement for
the airfoils in the different stations. Although a value decrease of aerodynamical
improvement, the hydrodynamical operation was benefitedt significantly because
the cavitation phenomena was avoided. These results demonstrate the feasibility
and effectiveness of the methodology used to improve the amphibious operation
in a 50-50 relationship eliminating adverse main effects during both operations.
96 C. Genaro Carreon et al.

References
1. Arias-Montano A, Coello CAC, Mezura-Montes E (2011) Evolutionary algorithms
applied to multi-objective aerodynamic shape optimization. In: Koziel S, Yang XS
(eds) Computational optimization, methods and algorithms. Springer, Heidelberg,
pp 211–240
2. Kulfan BM (2008) Universal parametric geometry representation method. J Aircr
45(1):142–158
3. Kulfan BM (2007) Recent extensions and applications of the “CST” universal para-
metric geometry representation method. In: AIAA paper
4. Kulfan BM, Bussoletti JE (2006) Fundamental parametric geometry representations
for aircraft component shapes. In: 11th AIAA/ISSMO multidisciplinary analysis and
optimization conference
5. Deb K, Pratap A, Agarwal S, Meyarivan TAMT (2002) A fast and elitist multiob-
jective genetic algorithm: NSGA-II. IEEE Trans Evol Comput 6:182–197
6. Drela M (1989) XFOIL: an analysis and design system for low Reynolds number
airfoils. In: Mueller TJ (ed) Low reynolds number aerodynamic, vol 54. Lecture
notes in engineering. Springer, Heidelberg, pp 1–12
7. Hájek J (2009) The generalized PARSEC parameterization for airfoil optimization.
In: Aerodynamic Research and Test Institute, Prague, 9 October 2009
8. Ponce LAJ, Montaño AA (2016) Parametrización Geométrica y Optimización
Aerodinámica de Alas en Flujo Transónico. Instituto Politécnico Nacional, Novem-
ber 2016
9. Yang X, Wang T, Liang J, Yao G, Liu M (2015) Survey on the novel hybrid
aquatic-aerial amphibious aircraft: aquatic unmanned aerial vehicle (AquaUAV). In:
ELSEVIER progress in aerospace sciences, pp 131–151
 Trajectory Tracking of Multiple
Quadrotors While Maintaining
Time-Varying Spatial Formations
via Synchronization

C.-A. Arteaga-Escamilla(B) , R. Castro-Linares(B) , and J. Álvarez-Gallegos(B)

Research Center for Advanced Studies (Cinvestav), Mexico City, Mexico

{cruzantonio.arteaga,rcastro,jalvarez}@cinvestav.mx

Abstract. In this paper, the trajectory tracking problem of multi-

ple quadrotors while maintaining time-varying spatial formations is
addressed through a synchronization approach. The aim is to control
each quadrotor to track its desired trajectory while its motion is syn-
chronized with the others within the formation to keep a desired time-
varying formation. First, a way to linearize the translational dynamics
of the quadrotor through feedback state is shown, considering the non-
linear dynamic model. An attitude PD controller for each quadrotor is
also designed to guarantee the attitude error approaches zero asymp-
totically. Second, a synchronous controller for each quadrotor transla-
tion is developed to guarantee that both position and synchronization
errors approach zero asymptotically. Finally, simulations were performed
to demonstrate the effectiveness of the proposed synchronization control
approach in the formation control tasks.

Keywords: Formation · Multiple quadrotors · Synchronization

1 Introduction

Study on the formation control of multiple quadrotors (Unmanned Aerial Vehi-

cles, UAVs for short) has received increasing attention in the last decade. There
are many strategies to solve the multiple UAVs formation control problem,
for example, leader-follower approach, virtual structure approach, cluster space
strategy, among others. In the leader–follower approach some robots take the
role of leaders and aim to track predefined trajectories, while the follower robots
follow the leaders according to a relative posture [6–8,13]. A disadvantage is that
there is no feedback from the followers to the leaders. In the virtual structure
approach, each robot is treated as an element of a rigid virtual structure, each of
them has a local control that makes the structure possible to maintain a given
arrangement [10,15,16]. This formation scheme is difficult to be implemented in

Supported by Conacyt, under Project CB-254329.

c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 97–108, 2020.
https://doi.org/10.1007/978-3-030-45402-9_11
98 C.-A. Arteaga-Escamilla et al.

some applications. The cluster space strategy conceptualizes the n-robot system
as a single entity (cluster), and desired motions are specified as a function of
cluster attributes (state variables) such as position, orientation, and geometry.
These state variables form the system cluster space. Cluster space state variables
may be related to robot-specific state variables through a formal set of kinematic
transformations. These transformations allow cluster commands to be converted
to robot specific commands to achieve the individual trajectory tracking [1,2,4].

In this paper we propose to use a synchronization control strategy to address

the multiple quadrotors formation control problem. The main idea is that a
team of quadrotors tracks each individual desired trajectory while synchronizing
motions among the others to keep relative kinematics relationship for maintain-
ing the desired time-varying shape. The main contribution of this formation
scheme is that the desired shape of the formation can take any spatial shape,
unlike [11,12,14,17,18], where the desired shape of the formation must have a
planar ellipse form. Besides, this formation scheme is easy to be implemented in
applications where the formation is spatial.

2 Dynamic Model of Quadrotor

Consider an inertial coordinate frame G = {XG , YG , ZG } fixed to ground and a
body fixed coordinate frame B = {XB , YB , ZB }. A picture of a quadrotor with
the corresponding frames is shown in Fig. 1. The generalized coordinates are
 T  T
given by ξ = x y z and η = φ θ ψ , where ξ denotes the spatial position
coordinate of the center of gravity respect to the inertial frame G, and η denotes
the Euler angles. These angles have the following names and they represent a
rotation around their corresponding axes: φ is the roll angle around XB axis,
θ is the pitch angle around YB axis, and ψ is the yaw angle around ZB axis.
The Newton-Euler approach was used to obtain the dynamical model of the
quadrotor, which considers a quadrotor as a rigid body under external forces
applied to the center of mass; so the dynamic model of the quadrotor is given
by [5,9].

mξ̈ = RB F − mg G D, (1)
J Ω̇ = −Ω × J Ω + τ , (2)
 T
where m is the mass of the quadrotor, G D = 0 0 1 is a unit vector in ZG
direction, g is the gravity constant; J ∈ IR3×3 denotes the inertia matrix respect
 T
to B; Ω = Ω1 Ω2 Ω3 represents the angular velocity in the body frame
 T
B. τ = τφ τθ τψ is the control torque defined in the body fixed frame B.
B
 T
F = 00u is the control input vector defined in the body fixed frame B.
4
u > 0 is the total thrust force generated by the rotors, this is u = i=1 fi ,
where fi is the thrust force generated by the ith rotor. R ∈ SO(3) is the so-
called rotation matrix that represents the orientation of the rigid body respect
 Trajectory Tracking of Multiple Quadrotors via Synchronization 99

to the inertial frame such that R : B → G is orthogonal, which is obtained by

making 3 successive rotations using the convention XB YB ZB . Using the compact
notation cθ = cosθ and sθ = sinθ, R is given by
⎡ ⎤
cθcψ sφsθcψ − cφsψ cφsθcψ + sφsψ
R = ⎣ cθsψ sφsθsψ + cφcψ cφsθsψ − sφcψ ⎦ (3)
−sθ sφcθ cφcθ

f2 f1
ZB
u
YB θ ψ XB
O
f3 f4
B
ZG

YG

G XG

Fig. 1. Coordinate systems and moment/forces acting on the quadrotor.

The kinematic relation of the angular velocity Ω and the generalized angular
velocities η̇, in the region where the Euler angles are valid, is given by [5,9].

Ω = Wη η̇, (4)

where ⎡ ⎤
1 0 −sinθ
Wη = ⎣ 0 cosφ sinφcosθ ⎦ (5)
0 −sinφ cosφcosθ
Substituting (4) in (2), the dynamic model of the quadrotor is obtained,
expressed in terms of the generalized coordinates, given by

mξ̈ = RB F − mg G D, (6)

J Wη η̈ = −J Ẇη η̇ − Wη η̇ × J Wη η̇ + τ , (7)
 T  T
with the control input vectors given by B F = 0 0 u and τ = τφ τθ τψ .
100 C.-A. Arteaga-Escamilla et al.

3 Control Strategy
Assuming that the closed-loop dynamics of rotation is much faster than the
translational one, it is possible to separate the model in two systems [3]. The
control strategy block diagram for each UAV is shown in Fig. 2, where the super-
script denotes the desired value of any variable. The control strategy consists of
the designing of a controller via synchronization for the translational dynamics
such that it guarantees the trajectory tracking and solves the formation control
problem, providing as its output the desired orientation to be fed as input to
the attitude stabilization controller. Such synchronous controller is used for all
UAVs within the formation.

ψd
B
ξ
d
ν F d d
O ,θ τ η
Synchronous Feedback Variables Attitude UAV
controller state conversion control
u ξ

Fig. 2. Control strategy block diagram for each UAV.

3.1 Dynamic Model Linearization

In order to linearize the translational dynamics of the UAV, consider (6), suppose
that it is possible to apply forces in all directions, i.e., the translational dynamics
of the UAV is given by
mξ̈ = RB T − mg G D, (8)
B
where T is a virtual control input vector. By using a feedback state given by
B
T = mR−1 (g G D + ν), (9)

where ν ∈ IR3 is an auxiliary control input vector, which will be defined in

Sect. 4; a linear translational dynamics is obtained

ξ̈ = ν.

Remark 1: Note that R is singular when cosφcosθ = 0.

Obviously, it is not possible to apply forces in all directions (only in ZB

direction). In order to B F force to have the same magnitude, direction and
sense of B T force, then the UAV must rotate, for that reason, in (6), R will be
replaced by Rd , which denotes the desired orientation of the UAV. Therefore,
next equations must be satisfied:

Rd (B F ) = RB T , (10)
 Trajectory Tracking of Multiple Quadrotors via Synchronization 101

and
B F =B T  . (11)
From (11), it is obtained
| u |=B T  .
Since the thrust force generated by the rotors is always positive, then the
thrust force that the rotors must generate is given by

u =B T  . (12)

Substituting (9) in (10), it is obtained

Rd (B F ) = m(g G D + ν)
⎡ ⎤
sψ d sφd + cψ d sθd cφd
⎣ −cψ d sφd + sψ d sθd cφd ⎦ u = m(g G D + ν)
cθd cφd
⎡ ⎤ ⎡ ⎤
sψ d sφd + cψ d sθd cφd G Nx
⎣ −cψ d sφd + sψ d sθd cφd ⎦ = m(g D + ν)
= ⎣ Ny ⎦ (13)
u
cθd cφd Nz
Taking ψ d as a constant value, where

ψ d ∈ [−π, π]\{±π/2},

it is possible to find the angles that the UAV must rotate to satisfy (10), these
angles are given by

φd = arcsin(Nx sinψ d − Ny cosψ d ), (14)

Nx − sinφd sinψ d
θd = arcsin . (15)
cosφd cosψ d
In order to linearize the rotational dynamics of the UAV consider (7), where
τ is the control input vector. By using a feedback state given by

τ = J Ẇη η̇ + Wη η̇ × J Wη η̇ + J Wη Γ , (16)
 T
where Γ = Γφ Γθ Γψ is an auxiliary control input vector, a linear rotational
dynamics is obtained and given by

η̈ = Γ . (17)

Remark 2: The dynamics of translation and rotation of the UAV are not inde-
pendent; because when applying an input torque given by τ , a rotation occurs
in the UAV, and this causes a resultant thrust force that will cause the UAV to
accelerate in XG or YG direction.
102 C.-A. Arteaga-Escamilla et al.

3.2 Attitude PD Controller

The attitude error of the UAV is defined as

η̄ = η d − η,
 T
where η d = φd θd ψ d is the desired attitude and η̇ d = η̈ d = 0. The proposed
attitude control law for the UAV is given by

Γ = Kdo η̄˙ + Kpo η̄ (18)

where Kpo , Kdo > 0. Substituting (18) in (17), it is obtained an attitude error
dynamics of the UAV given by

¨ + Kdo η̄˙ + Kpo η̄ = 0,

η̄

which is asymptotically exponentially stable.

4 Synchronous Controller Design

Consider a group of n UAVs with their translational dynamics given by

ξ̈ i = ν i , (19)
 T
where ξ i = xi yi zi denotes the spatial position coordinate of the ith UAV
and ν i is the control input respect to ξ i . The position tracking error of the ith
UAV is defined as
ei = ξ di − ξ i , (20)
where ξ di is the desired spatial position coordinate. Similar to [14], the synchro-
nization error of the ith UAV is defined as

i = ei − ei+1 . (21)

Note that when i = n, i + 1 is denoted as 1. Besides, if the synchronization error

i = 0 for all i = 1, 2, ..., n, then the synchronization goal is achieved.

In a similar manner to [14], in order to link the position and synchronization

errors into one, the coupling error of the ith UAV is defined as
t
E i = ei + β (i − i−1 )dζ, (22)
0

where β > 0. Note that the coupling error feeds back the information of the two
neighboring UAVs i − 1 and i + 1. Note that when i = 1, i − 1 is denoted as n.
Differentiating E i with respect to time yields

Ė i = ėi + β(i − i−1 ), (23)

 Trajectory Tracking of Multiple Quadrotors via Synchronization 103

Ëi = ëi + β(˙i − ˙i−1 ). (24)

The proposed control law for each UAV is given by
d
ν i = ξ̈ i + β(˙i − ˙i−1 ) + ΛĖ i + Υ E i , (25)

where Λ, Υ > β. Substituting the controller (25) into the UAV translational
dynamics in (19) leads to the coupling error dynamics given by

Ë i + ΛĖ i + Υ E i = 0, (26)

which is asymptotically exponentially stable. Therefore, the proposed syn-

chronous controller (25) leads to asymptotic exponentially convergence of both
position tracking and synchronization errors to zero.

5 Simulation

A simulation was performed to verify the effectiveness of the proposed synchro-

nization control approach.

Initial and final shape

Q20

0.2 Initial Shape

Final Shape

0.15
Z [m]

0.1

0.05

0 Q2f
1
Q10
Q 1
0.5
Q30 Q3f 0.5 1f
0
-0.5
Y [m] 0 -1
X [m]

Fig. 3. Initial and final desired shape.

5.1 Build the Formation, Trajectory Tracking and Switch of Shapes

Simulation is divided in three parts, every part takes 10 s. First part consists of
each UAV reaches its desired position to build the desired initial shape. The sec-
ond part consists of trajectory tracking while maintaining the formation shape.
Finally, the UAVs have to switch from the initial to the final desired shape.
Figure 3 illustrates the initial and final desired shape, where each vertex rep-
resents each UAV. The desired shape is defined by specifying distances among
104 C.-A. Arteaga-Escamilla et al.

Spatial positions

UAV1
1.3
UAV2
UAV3
1.2

Z [m]
1.1

0.9
3

2
30
1
20
0 10
0
Y [m] -1 -10 X [m]

Fig. 4. Spatial positions of UAVs.

UAVs or specifying desired position coordinates respect to a reference point

(the origin is the reference point). To build the desired initial shape, the desired
coordinates of each UAV are: Q10 (1, 0, 1), Q20 (0, 1, 1.2) and Q30 (−1, 0, 1). For
the desired final shape, the desired coordinates of each UAV are: Q1f (0.7, 0, 1),
Q2f (0, 0.5, 1) and Q3f (−0.7, 0, 1), as shown in Fig. 3. L1 denotes the distance
between UAV 1 and 2, L2 denotes the distance between UAV 2 and 3, and L3
denotes the distance between UAV 3 and 1. The desired trajectory parameters
are: xd (t) = t, y d (t) = sin(wt), z d (t) = 1, with w = 2π/40. The distance and
time are expressed in meters and seconds, respectively.

Positions in plane XY
2.5
UAV1
UAV2
2 UAV3

1.5
Y [m]

0.5

-0.5
-5 0 5 10 15 20 25
X [m]

Fig. 5. Positions in plane XY of UAVs.

 Trajectory Tracking of Multiple Quadrotors via Synchronization 105

Position tracking error (e)

0.6
e 1x
e
1y
0.4 e 1z
e 2x
e
2y
0.2
e 2z

Tracking error [m]

e 3x

0 e 3y
e 3z

-0.2

-0.4

-0.6
0 5 10 15 20 25 30
time [s]

Fig. 6. Position tracking error throughout the time.

Position tracking error (e)

0.6
e 1x
e
1y
0.4 e 1z
e 2x
e
2y
0.2
e 2z
Tracking error [m]

e 3x

0 e 3y
e 3z

-0.2

-0.4

-0.6
0 5 10 15 20 25 30
time [s]

Fig. 7. Coupling error throughout the time.

All UAVs are identical, whose parameters are: mass m = 100 g, maximum
thrust force umax = 1.05N, maximum input torque τφmax = τθmax = τψmax =
1×10−3 Nm, inertia matrix J = diag(2.916×10−4 , 3.075×10−4 , 5.707×10−4 ) kg∗
2
m2 , gravity constant g = 9.81 m/s . All UAVs start from hover with the next
initial positions and orientations: ξ10 = [ 0.7 0 0.9 ]T , ξ20 = [ 0.5 0.7 1 ]T , ξ30 =
[ −0.8 0.1 0.9 ]T , η10 = η20 = η30 = [ 0o 0o 0o ]T . The desired yaw angles are:
ψ1d = ψ2d = ψ3d = 0o . Control gains are: kpo = 2, kdo = 0.5, β = 0.5, Λ = 2.2,
Υ = 1.6.

Figure 4 illustrates the spatial positions of the UAVs throughout the time.
Figure 5 illustrates the positions of the UAVs in plane XY throughout the time.
It is seen that there is an impulse when switching between shapes. Figures 6 and
7 depict, respectively, the time evolution of the position tracking and coupling
106 C.-A. Arteaga-Escamilla et al.

Sychronization error ( )
0.8

0.6 1x

1y

0.4 1z

Synchronization error [m]

2x

2y
0.2
X: 7.915 2z
Y: 2.238e-06 3x
0
3y

3z

-0.2

-0.4

-0.6
0 5 10 15 20 25 30
time [s]

Fig. 8. Synchronization error throughout the time.

Distances among UAVs (L, Ld )

2.2
L1
2 L2
L
3
1.8 Ld1
Distances among UAVs [m]

Ld2
1.6
Ld
3

1.4

1.2

0.8

0.6
0 5 10 15 20 25 30
time [s]

Fig. 9. Real and desired distances among UAVs throughout the time.

error. It is seen that both errors increase their value at the beginning when
building the formation, when trajectory tracking (at 10 s), and when switching
between shapes (at 20 s), but both errors converge to zero successfully after 6 s
approximately. Figure 8 depicts the synchronization error, which increases its
value at the beginning and when switching between shapes (at 20 s), but not
when trajectory tracking starts (at 10 s), unlike position and coupling errors
shown in Figs. 6 and 7, respectively. Figure 9 depicts the time evolution of the
real and desired distances among UAVs. It is seen that desired distances are
reached after 8 s approximately, and they maintain their value during trajectory
tracking from 10 to 20 s. Finally, final desired distances are reached 7 s after
switching of shapes.
 Trajectory Tracking of Multiple Quadrotors via Synchronization 107

6 Conclusions and Future Work

In this paper, a synchronous controller is developed for switching between for-
mations and for trajectory tracking of multiple quadrotors while maintaining
spatial formations. A synchronization error is defined as differential position
error between every pair of two neighboring UAVs. A coupling error is defined
to link the position and synchronization errors. It is shown that the developed
controller guarantees asymptotic convergence to zero of both position and syn-
chronization errors. Simulation results show the effectiveness of the proposed
synchronous controller. It is seen that this controller is not robust, i.e, it is not
possible to eliminate effects caused by external disturbances or by parametric
uncertainties. Future work will be to add robustness to the proposed controller
and its implementation in a real experimental platform.

References
1. Agnew MS (2009) Cluster space control of a 2-robot aerial system. Doctoral dis-
sertation, Santa Clara University, School of Engineering
2. Agnew MS, Dal Canto P, Kitts CA, Li S (2010) Cluster space control of aerial
robots. In: 2010 IEEE/ASME international conference on advanced intelligent
mechatronics, pp 1305–1310
3. Bertrand S, Guénard N, Hamel T, Piet-Lahanier H, Eck L (2011) A hierarchical
controller for miniature VTOL UAVs: design and stability analysis using singular
perturbation theory. Control Eng Pract 19(10):1099–1108
4. El Ferik S, Nasir M-T, Baroudi U (2016) A Behavioral Adaptive Fuzzy controller
of multi robots in a cluster space. Appl Soft Comput 44:117–127
5. Garcı́a-Carrillo L-R, Dzul-López A-E, Lozano R, Pégard C (2012) Quad rotorcraft
control. Vision-based hovering and navigation. Springer, London
6. Guo C, Qin J, Ge Y, Chen Y (2014) Leader-follower and cascade system based
formation control of nonholonomic intelligent vehicles. In: The 26th Chinese control
and decision conference (CCDC 2014). IEEE, pp 3931–3935
7. Mercado DA, Castro R, Lozano R (2013) Quadrotors flight formation control using
a leader-follower approach. In: 2013 European control conference (ECC). IEEE, pp
3858–3863
8. Peng T, Li S (2014) Formation control of multiple wheeled mobile robots via leader-
follower approach. In: The 26th Chinese control and decision conference (CCDC
2014). IEEE, pp 4215–4220
9. Quan Q (2017) Introduction to multicopter design and control. Springer, Singapore
10. Sadowska A, den Broek T-V, Huijberts H, van de Wouw N, Kostić D, Nijmeijer H
(2011) A virtual structure approach to formation control of unicycle mobile robots
using mutual coupling. Int J Control 84(11):1886–1902
11. Sanhoury I-M, Amin S-H, Husain A-R (2012) Multiple nonholonomic wheeled
mobile robots trajectory tracking while maintaining time-varying formation via
synchronous controller. Procedia Eng 41:1044–1050
12. Sanhoury IM, Amin SH, Husain, AR (2012) Switching between formations for
multiple mobile robots via synchronous controller. In: 2012 IEEE 8th international
colloquium on signal processing and its applications, pp 352–357
108 C.-A. Arteaga-Escamilla et al.

13. Soorki MN, Talebi HA, Nikravesh SKY (2011) A robust dynamic leader-follower
formation control with active obstacle avoidance. In: 2011 IEEE international con-
ference on systems, man, and cybernetics. IEEE, pp 1932–1937
14. Sun D, Wang C, Shang W, Feng G (2009) A synchronization approach to trajec-
tory tracking of multiple mobile robots while maintaining time-varying formations.
IEEE Trans Rob 25(5):1074–1086
15. Tan KH, Lewis MA (1996) Virtual structures for high-precision cooperative mobile
robotic control. In: Proceedings of IEEE/RSJ international conference on intelli-
gent robots and systems, IROS 1996, vol 1, pp 132–139
16. Van den Broek TH, van de Wouw N, Nijmeijer H (2009) Formation control of
unicycle mobile robots: a virtual structure approach. In: Proceedings of the 48th
IEEE conference on decision and control (CDC) held jointly with 2009 28th Chinese
control conference. IEEE, pp 8328–8333
17. Wang C, Sun D (2008) A synchronous controller for multiple mobile robots in
time-varied formations. In: 2008 IEEE/RSJ international conference on intelligent
robots and systems. IEEE, pp 2765–2770
18. Zhao D, Zou T (2012) A finite-time approach to formation control of multiple
mobile robots with terminal sliding mode. Int J Syst Sci 43(11):1998–2014
 Modeling, Simulation and Analysis
of the Characteristics of the Non-conventional
Vertical Axis Wind Turbine with Variable Pitch
Angle that Allows to Establish Different
Control Strategies

Jesús García-Barrera1,2(B) , Jorge Díaz-Salgado1(B) , Oliver M. Huerta-Chávez1(B) ,

and Rafael Sánchez-López1(B)
1 Tecnológico de Estudios Superiores de Ecatepec, 55210 Ecatepec de Morelos,
Edo. De México, Mexico
j3susgb90@gmail.com, {jorgesalgado,rsanchez}@tese.edu.mx,
lecram_21@hotmail.com
2 Tecnológico de Estudios Superiores de Tianguistenco, 52650 Tianguistenco,

Edo. De México, Mexico

Abstract. This work presents the modeling, simulation and analysis in stationary
response of the principal characteristics of a vertical axis wind turbine (VAWT)
with variable pitch angle in order to establish different control techniques. The
mathematical static model implemented for the study is the so called Double
Multiple Stream Tube. For different initial conditions, the model implementation
in a specialized software permits to obtain: (i) position vs output torque plots of the
wind turbine and of each blade, and (ii) 3D angular position vs average moment
vs attack angle plots. The analysis of these plots permits the establishment of an
adequate pitch angle at each angular position (from 0 to 2π) for the improvement
of the output generated torque of the wind turbine.

Keywords: VAWT · Variable pitch angle · Numerical simulation · Double

multiple stream tube model · Momentum models

1 Introduction

Actually, wind energy is exploited by means of wind parks conveniently placed in high
wind resource rural areas. Commonly these parks are a set of horizontal wind turbines
positioned at high elevation from the ground to reduce the air turbulence. The Horizontal
Axis Wind Turbine (HAWT) technology is mature and commercially exploited in this
wind parks. This is very far in the case of Vertical Axis Wind Turbine (VAWT) since
a lot of technological developments and improvements are necessary. Despite a lower
efficiency from respect to the HAWT, VAWT mechanisms offers, by the way, many
advantages: low starting torque, omnidirectionality, high turbulence operating capability,

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 109–118, 2020.
https://doi.org/10.1007/978-3-030-45402-9_12
110 J. García-Barrera et al.

and low cost [1, 2]. These features make these turbines to be ideal to be used in urban
areas with low wind resource [3].
The reported works of VAWT turbines have been primary concern to: (i) the analysis
of different aerodynamical airfoils (ii) the proposition of different mathematical models
to predict the behavior of the turbine in steady state, (iii) the calculus of theoretical
efficiency based in these proposed models, the number of blades and the type of aero-
dynamical airfoil and (iv) the effect of blade pitch angle with active and passive control
on aerodynamic performance [4–7].
In this work, a static model of a two blade VAWT is proposed and simulated in a
specialized numerical software. The static model of a wind turbine is a fundamental
tool for the analysis and design of this mechanism, with that in mind here the modeling,
simulation and analysis of the behavior of a VAWT is done. The model permits to
simulate the effect of a variable pitch angle with a desired trajectory along a rotation
cycle.
The methodology followed consists in: (i) the proposition of a Double Multiple
Stream Tube mathematical model, (ii) the implementation of this mathematical model
in a numeric simulation software, (iii) the realization of various simulations in order
to establish a suitable pitch angle to improve the output torque, and (iv) variable pitch
simulations to determine the turbine behavior with these proposed trajectory. Section 2
presents the mathematical model and its characteristics, Sect. 3 deal with the implemen-
tation of the model. In Sect. 4 the simulated results are presented and discussed. Finally
some conclusions are drawn in Sect. 5.

2 Mathematical Model
The theoretical analysis of the VAWT is based in the so called Double Multiple Stream
Tube mathematical model. This model has been widely used to predict the total torque
and aerodynamic forces of a VAWT with straight or curved blades [8]. The model is
based in other two models: (i) two actuator disc model, and (ii) multiple stream tube
model.

2.1 Two Actuator Disk Model

This model is based in the partition of the turbine influence zone in three parts: (i) the
upwind high velocity Ui zone, (ii) an equilibrium velocity Ue zone and a (iii) downwind
low velocity Ui zone (Fig. 1).
Equations (1)–(4) describe the induced velocities in each zone and the velocity
induction factor (2) [9].
Ui = U∞ (1 − a) (1)
U∞ − Ue
a= (2)
U∞
Ue = U∞ (1 − 2a) (3)
 
Ui = Ue 1 − a  (4)
 Modeling, Simulation and Analysis of the Characteristics 111

Fig. 1. Two actuator disc model.

2.2 Multiple Stream Tube Model

This model [10] permits to compute the mean turbine torque considering multiple stream
tubes along the turbine influence zone parallel to the wind direction. Each considered
stream tube has its own interference velocity (Fig. 2). The accuracy of the model depends
of the number of streams taken into account.

Fig. 2. Multiple stream tube model.

2.3 Double Multiple Stream Tube Model

The Double Multiple Stream Tube Model (Fig. 3) is the combination of the two previous
models. Considers a two actuator disc model for the high velocity zone and another for
the low velocity one, in this way two actuator discs are located in series. Using Eqs. (5),
(6), (7) it is possible to calculate the induced velocities in each streams tube and zones
(Ui , Ue , Ui ).

Ui = U∞ a (5)
112 J. García-Barrera et al.

Ue = U∞ (2a − 1) (6)

Ui = Ue a  (7)
 π
N 2
f up = |sec(θ )|(Cn cos(θ ) − Ct sen(θ ))dθ (8)
8π R − π2

 3π
N 2
f dw = |sec(θ )|(Cn cos(θ ) − Ct sen(θ ))dθ (9)
8π R π
2
π
a= (10)
f up + π
π
a = (11)
f dw + π
1
f nup(θ) = ρcL W 2 Cn (12)
2
1
f tup(θ) = ρcL W 2 Ct (13)
2
1
Q up(θ) = ρc R LCt W 2 (14)
2
 π
N 2
Q AV U P = Q up(θ) dθ (15)
2π − π2
 3π
N 2
Q AV DW = Q dw(θ) dθ (16)
2π π
2

Fig. 3. Double multiple stream tube model.

 Modeling, Simulation and Analysis of the Characteristics 113

Equations (10) and (11) permit to calculate low and high velocity induction factors
for each stream tube (a, a  ) by iterating Eqs. (8) and (9). Normal f nup(θ) and tangential
f tup(θ) forces and output torque Q up(θ) for each angular position are calculated by
Eqs. (12), (13) and (14). Finally, the high Q AV U P and low Q AV DW velocity average
torque can be obtained by Eqs. (15) and (16).

3 Model Software Implementation

The proposed model was implemented in a specialized software for numerical simula-
tion. The implementation is based in the reported iterative algorithm presented in [11].
This iterative algorithm computes: (i) the average torque, (ii) tangential forces and (iii)
normal forces that the turbine generates for each angular position in steady state. In
order to get those results the algorithm must satisfy a convergence criteria. This criteria
consist in an error defined as the difference between the actual and previous interference
factor. Nevertheless, the reported algorithm lacks an interpolate subroutine to calculate
the lift and drag coefficients of the turbine. The problem was surpassed with: (i) the
addition of an interpolation method and (ii) the inclusion of experimental data tables of
the considered NACA0012 airfoil [12]. Another improvement of the original algorithm
is the incorporation of a variable attack-pitch angle.
The modified algorithm permits to: (i) test and analyze the static behavior with
variable attack-pitch angle, and (ii) generate curves of the angular position-output torque
and angular position-attack angle. These improvements permit to analyze the average
output torque with a variable pitch angle with fixed T.S.R.
The proposed algorithm (Fig. 4) needs three input parameters: (i) air velocity V∞ ,
(ii) angular velocity ω and (iii) VAWT radius R in order to obtain the average output
torque.

Fig. 4. Implemented static algorithm block diagram.

Employing these input parameters, the proposed algorithm computes the average
torque by a bi-dimensional interpolation that relies on: Reynolds number, relative veloc-
ity and experimental lift and drag coefficients. For each stream tube an iterative process
computes: (i) the interference factors a y a , (ii) Reynolds number, (iii) attack-pitch
angle, and (iv) normal and tangential forces.
114 J. García-Barrera et al.

The drawbacks of the proposed method are: (i) the algorithm requires an iterative
process in order to compute the interference factor for each stream tube and (ii) the
computed average torque is the steady state value response.

4 Simulation Results
Four different speed velocities from 5 to 20 m/s are simulated with the parameters
depicted in Table 1 and conditions of Table 2 with a constant T.S.R. of 4.2.

Table 1. Physical parameters.

Parameter Value Units

Blade number 2 Unity
Blade chord 0.3 m
Blade length 2.0 m
Turbine radius 2.0 m

Table 2. Simulation conditions.

Simulation conditions Test 1 Test 2 Test 3 Test 4

Attack angle 0º [rad] 0º [rad] 0º [rad] 0º [rad]
Number of stream-tubes 20 20 20 20
Wind velocity [m/s] 5 10 15 20
Angular velocity [rad/s] 10.5 21 31.5 42
T.S.R. [/] 4.2 4.2 4.2 4.2
Turbine solidity 0.15 0.15 0.15 0.15

The angular position θ vs output torque (Fig. 5), and angular position vs attack angle
vs output torque of each blade (Fig. 6) is showed for each air velocity.
These results show that the tendency of the curves is similar for different air velocities,
according with the theory, a quadratic relationship between output torque and air velocity
is evident (Figs. 5 and 6). As expected, the maximum torque appears in the high velocity
zone. In particular this occurs between 0º and 20º that corresponds to an attack angle
between 9º and 10º (Fig. 6). Meanwhile, the minimum torque arises in an angular position
around 270º that corresponds to an attack angle between −2º and 0º.
The steady state simulated results permit to propose that a constant attack angle equal
to 10° will upgrade the output torque. Test 1 is repeated with this constant attack angle
(Fig. 7).
 Modeling, Simulation and Analysis of the Characteristics 115

Fig. 5. Angular position vs output torque.

Fig. 6. Rotation angle vs attack angle vs output torque 3D plot.

Fig. 7. Difference between the angular position and free and constant attack angle for Test 1.
116 J. García-Barrera et al.

The output torque with and without constant attack angle shows a significant output
torque improvement of about four times (≈3.75) the original output (Fig. 8).

Fig. 8. Difference between the output torque generated with free and constant attack angle.

The respective desired pitch angle (for a constant attack angle of 10°) is presented in
Fig. 9, and establishes the desired trajectory of each blade along a revolution cycle. The
tracking of this pitch angle trajectory would imply a technical challenge, and establishes
a first comparison with a theoretical value that significantly improves the output torque.

Fig. 9. Pitch angle required to maintain a constant attack angle.

5 Conclusions
The Double Multiple Stream Tube mathematical model was successfully implemented
in a numerical simulation software to shape the attack angle and output torque of a two
blade vertical axis wind turbine (VAWT) with variable pitch angle.
 Modeling, Simulation and Analysis of the Characteristics 117

The implemented model permits to obtain: (i) angular position vs output torque plots
of the wind turbine and of each blade, and (ii) 3D angular position vs average moment vs
attack angle plots. The steady state simulated results permit to conclude that the output
torque can be upgraded with an appropriated attack angle. Moreover, the analysis of
these plots permits to establish a desired pitch angle trajectory at each angular position
(from 0 to 2π) for the improvement of the average generated torque of the wind turbine.
The proposed pitch angle trajectory was obtained considering a constant attack angle
specified near the maximum output torque. This yields a theoretical significant out-
put torque improvement of about four times the original output. Indeed, the desired
pitch angle trajectory would imply a great technical challenge, and its implementation
would be a difficult task, it establishes a first comparison with a theoretical value that
significantly improves the output torque.
There is a lot of additional work to be done: (i) first the experimental implementation
of the desired pitch angle trajectory, (ii) the consideration of different T.S.R.’s and blade
solidity, (iii) greater air velocities with turbulent flow, (iv) a greater number of blades,
and (v) the consideration of a symmetrical central blade, normally used in this types
of turbines to automatic start up. Finally, the numerical model must include a greater
number of table entries for the Reynolds number computing in order to outperform the
numerical precision.
Meanwhile with the desired pitch angle also many different trajectories and control
strategies can be proposed and tested as future work.

References
1. Musgrove PJ (1979) Vertical axis wind turbine. 0–5. https://doi.org/10.1016/b978-1-4832-
8407-1.50354-8
2. Aslam Bhutta MM, Hayat N, Farooq AU, Ali Z, Jamil SR, Hussain Z (2012) Vertical axis
wind turbine - a review of various configurations and design techniques. Renew Sustain.
Energy Rev 16:1926–1939. https://doi.org/10.1016/j.rser.2011.12.004
3. Chaichana T, Chaitep S (2010) Wind power potential and characteristic analysis of Chi-
ang Mai, Thailand. J Mech Sci Technol 24:1475–1479. https://doi.org/10.1007/s12206-010-
0415-3
4. Zhang LX, Liang YB, Liu XH, Guo J (2014) Effect of blade pitch angle on aerodynamic
performance of straight-bladed vertical axis wind turbine. J Cent South Univ 21:1417–1427.
https://doi.org/10.1007/s11771-014-2080-7
5. Zhang L, Pei Y, Liang Y, Zhang F, Wang Y, Meng J, Wang H (2015) Design and implementation
of straight-bladed vertical axis wind turbine with collective pitch control. In: 2015 IEEE
international conference on mechatronics and automation (ICMA), pp 1258–1263. https://
doi.org/10.1109/ICMA.2015.7237666
6. Zhang L, Liang Y, Li E, Zhang S, Guo J (2012) Vertical axis wind turbine with individual active
blade pitch control. In: Asia-Pacific power and energy engineering conference, APPEEC.
https://doi.org/10.1109/APPEEC.2012.6307108
7. Pawsey NCK (2002) Development and evaluation of passive variable-pitch vertical axis wind
turbines
8. Paraschivoiu I (2002) Wind Turbine Design With Emphasis on Darrieus Concept. École
Polytechnique de Montréal
118 J. García-Barrera et al.

9. Mohammed AA, Ouakad HM, Sahin AZ, Bahaidarah HMS (2019) Vertical axis wind turbine
aerodynamics: summary and review of momentum models. J Energy Resour Technol Trans
ASME 141:1–10. https://doi.org/10.1115/1.4042643
10. Paraschivoiu I (1988) Double-multiple stream tube model for studying vertical-axis wind
turbines. J Propuls Power 4:370–377
11. Castillo J (2011) Small-Vertical Axis Wind Turbine Design. 69
12. Sheldahl RE, Klimas PC (1981) Aerodynamic characteristics of seven symmetrical airfoil
sections through 180-degree angle of attack for use in aerodynamic analysis of vertical axis
wind turbines. Chem SAND 80-211, 118. https://doi.org/10.2172/6548367
 Numerical Simulation of the Formation,
Evolution and Control of a Methanol Spray
Generated by a Pressure Swirl Atomizer During
the Combustion Process

G. Guevara-Morales1,2 , O. M. Huerta-Chavez1,2(B) , A. Arias-Montaño1,2 ,

M. Zirion Flores1,2 , and I. Castorena Solache2,3
1 SEPI-ESIME Ticomán-IPN, Gustavo A. Madero, CDMX, Mexico
Lecram_21@hotmail.com
2 Catedras Conacyt TESE Posgrado de Mecatronica, Santiago de Querétaro, Mexico
3 ISAE ENSMA, Chasseneuil du Poitou, France

Abstract. In this study the injection of liquid fuel by a pressure swirl atomizer
during the combustion process, is reproduced and analyzed using numerical sim-
ulation (Computational Fluid Dynamic). A 2D model along with an Eulerian-
Lagrangian method has been use to simulate the presence of the spray and its
interaction with the continuous phase. The injection and atomization of the fuel
are reproduce using the LISA (“Linearized Instability Sheet Atomization”) model
together with a Rosin-Rammler distribution, while the dispersion (evolution) of
the droplets is simulated through the governing equations of the discrete phase.
On the other hand, the continuous phase is simulated using a model based on the
Reynolds-averaged Navier-Stokes equations (RANS). The combustion model use
is the steady-flamelets. The results of the dispersion (number density), size (Sauter
mean diameter) and velocity of the droplets are compare with the experiments,
showing good agreement. Finally, in order to study the control of the injection
process, the operating parameters of the atomizer have been modified, analyzing
the response in the characteristics of the spray.

Keywords: Simulation · RANS · Injection and control

1 Introduction

For research and development, the combustion process, numerical tools such as CFD
are implemented. CFD tools can emulate the behavioral details of the flow inside the
turbulent combustion chamber without the need of sensors, therefore not invading and
not altering the behavior of the flow. It is then very useful to choose the experiments that
will be carried out based on the results that the model yields. This is possible thanks to
the development of models ever more precise and ever more complex use for the study
of the aforementioned phenomena, besides this can be done at a fraction of the cost of
the experimental setup.
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 119–137, 2020.
https://doi.org/10.1007/978-3-030-45402-9_13
120 G. Guevara-Morales et al.

Regarding the combustion modelling, the non-linear nature of the chemical reaction
rate (ω), expressed by an Arrhenius law, the mean ω cannot be described as function of
the means of the variables that define it (mass concentration, density and temperature).
Such being the case, the transport equations for different chemical species cannot be
expressed in the same way as the ones that deal with turbulence. To deal, different
models have been developed, based on relatively easy to model variables. Veynante and
Vervisch [3] feature an exhaustive revision of combustion models, these can be separated
into two categories depending on how they treat chemical rate: either an infinitely fast
chemical rate or a finite one. The main way to determine which type of model suits the
needs of one specific problem is to compare the chemical rate characteristic time (τc)
and the characteristic time of turbulence of the flow (τt). Damköhler’s number is meant
to characterize the relation between these times, as Da = τt/τc. If Da 1 then one can
consider an infinitely fast reaction and the turbulence will characterize the behavior of
the flow [3].
Models that consider infinitely fast chemical reactions assume that reagents are
burned as soon as the right proportions of mixture are reached. The chemical process
is fast compared to the transport process (diffusion and convection). As a result, the
knowledge of the mixing profile of the chemical species provides us with knowledge
on their oxidation and therefore combustion. Among these chemical equilibrium-driven
models in diffusion flames we find models that are based on determining mixture fraction
(f) inside the combustion chamber. (f) values ranges between 1 and 0 where 1 represents
a flow region in which only fuel is present and 0 represents a region in which only oxygen
is present. The rest of the field unknowns are determined by exploiting linear relations
between them and mixture fraction (f). The model used in this paper (flamelets) is based
on the mixture fraction (f) [3–5], it assumes that the turbulent flame is a superposition
of laminar flames that vary in position through time. It can be observed that the position
of the flame and reaction of the species is directly affected by the relative velocity
and distance between the oxidizer and fuel flows and we can clearly identify the regions
where the flame is present and its extinction once the mixing proportions do not allow for
a flame to form anymore. These flow characteristics are described in a variable called
scalar dissipation (χ). Because f varies along the distance between fuel and oxidizer
flows, χ is defined based on f’s gradient, therefore representing the shape of the flame.
The flamelets model takes this narrowing down of the flame into account thanks to the
aforementioned variables unlike a simple chemical equilibrium model. To couple the
combustion model with the turbulence model we have an Eddy-Breakup model that
considers that the reactions take place in the smaller scales of turbulence [3], in which
the molecular level mixing occurs. Given that we consider infinitely fast reactions, the
flow of reagents and the turbulent mixing process are the limiting process inside the
combustion chamber and determines the actual evolution of the flame.
The present paper a 2D model along with a Eulerian-Lagrangian method had been
used to simulate the presence of the spray and its interaction with the continuous phase.
The injection and atomization of the fuel are reproduced using the LISA (“Linearized
Instability Sheet Atomization”) model together with a Rosin-Rammler distribution,
while the dispersion (evolution) of the droplets is simulated through the governing equa-
tions of the discrete phase. On the other hand, the continuous phase is simulated using a
 Numerical Simulation of the Formation, Evolution and Control 121

model based on the Reynolds-averaged Navier-Stokes equations (RANS). The combus-

tion model used is the steady-flamelets. The results of the dispersion (number density),
size (Sauter mean diameter) and velocity of the droplets are compared with the exper-
iments, showing good agreement. Finally, in order to study the control of the injection
process, the operating parameters of the atomizer have been modified, analyzing the
response in the characteristics of the spray. The model will be evaluated when compared
to the experimental setup of combustion chamber carried out by the NIST (National
Institute of Standards and Technology) [1]. The experimental parameters and results
have been used to validate the data yielded by the coupled models.

2 Methodology
2.1 Experimental Setup
The NIST experimental along with the database by Widmann and Presser [1] were
selected as the reference values for the present article. The NIST model, as illustrated
in Fig. 1, consists of a cylindrical reservoir made of steel and kept at ambient pressure.
Inside the cylindrical container a diffusion flame is generated via the injection of liquid
methanol (CH3OH) as a spray and air with a swirling motion. The spray is created by
using a pressure injector which forms a hollow cone of fuel. Around this system there’s
a vortex generator (swirler) with 12 adjustable blades.
This experimental setup was used to create a database that could validate a com-
bustion model. The spray measurements were obtained in a non-intrusive fashion using
Phase Doppler Interferometry (PDI). The burnt gases temperature at the combustion
chamber’s outlet as well as the temperature of the chamber walls were measured by
thermocouple probes. To characterize the turbulent flow Particle Image Velocimetry
(PIV) was employed. The results reported in [1], measured during the stable status of
the process are listed below:

– Spray: Sauter’s diameter, droplet density, volumetric flow, and, axial and radial mean
velocities of drops. This set of data was measured multiple times at different altitudes
of the combustion chamber counting form the injection point with their respective
error margins. The altitudes were: (5, 15, 25, 35, 45, 55 and 65 mm from the injection
point.
– Flow: Axial, radial and tangent velocities as well as their fluctuations, measured in
function of their radial position and at three axial planes from the injection point (1.4,
9.5 and 17.6 mm) the data is reported when the spray is present as well as when it
isn’t. It also includes measurement error margin.
– Temperature: Mean temperature of the container’s Wall measured at 10 different
altitudes from its base (99, 165, 231, 297, 429, 561, 693, 826, 958 and 1090 mm).
Mean temperature of outlet gases measured at 13 different points distributed along
the outlet.

The NIST model was chosen as a starting point for the present paper mainly because
the experimental values were extensively reported and readily available alongside the
operating parameters for the experiment. This allows for easier definition of boundary
122 G. Guevara-Morales et al.

conditions in the simulations and for meaningful comparisons to validate the model.
Besides, this models can be simplified to focus computer resources on the analysis of
the fundamental phenomena intervening in turbulent combustion. The preceding results
and considerations of the RANS simulations based on this same NIST experimental
setup (Zhu et al. [10], Crocker et al. [11] and Collazo et al. [12]) form the angular stone
in which this simulation is built on.

Fig. 1. Experimental Model from NIST, units: mm [1].

2.2 Computer Simulation RANS 2D

As shown “Experimental Model (Fig. 1)”, the burnt gas outlet is placed at the side of the
combustion chamber, this makes the setup asymmetrical. However, according to Zhu
et al. [10] and Crocker et al. [11] the effect of having non-centered outlet is irrelevant to
the flow in the vicinity of the injector, zone that will be studied in this paper. This means
that the combustion chamber’s geometry can be simplified by considering the upper face
of the reservoir as the outlet for the gases, giving the combustion chamber axial symmetry
and making a 2D simplification of the chamber and flow possible. As a result, only half
of the mean plane of the combustion chamber is modelled. This has great effects in the
overall simulation time. Even though there are no tangential components in the gradients
of an axisymmetric flow, we must still consider the radial fluctuation in velocity of the
flow due to turbulence. In a spray’s particular case, keeping the fluctuation of velocity in
mind implies that injected particles and their properties are also axisymmetric. Although
the spray’s 3D simulation is able to emulate these fluctuations and the 2D is not, the
2D simulation is capable of rendering the general characteristics of the spray [10]. In
“Computational Model for RANS Simulation (Fig. 2), a 2D model for the combustion
chamber is shown, the inlet and outlet for the gases are outlined. The outlet’s area is the
same as in the experimental setup, with a 203.2 mm radius.
Taking into account the simple geometry of the combustion chamber a structured
mesh of 338 058 square elements was used, the mesh was refined around the air and fuel
 Numerical Simulation of the Formation, Evolution and Control 123

inlets. The mesh’s parameters were determined thanks to a mesh independence study.
In the aforementioned study the mean temperature in relation to the mass contained
in each of the mesh’s elements was compared between the elements. In contrast with
a classical transient simulation in which the solution converges to a steady, invariant
state, the coupling of the spray’s transient model modifies the results. In this case, the
injected particles’ properties into the chamber such as size, velocity, trajectory, position
and collision modes vary randomly during each time step. As a result, the liberated heat
coming from the chemical reaction varies and consequently the global temperature of the
combustion chamber varies too. As this is the case the simulation’s result will converge
to a statistic stability state instead of a steady value. In this state, the solution will be
in the vicinity of a predicted value (statistical mean) and will have a given variance.
Therefore, it is the statistical mean of the temperature obtained by the convergence of
the simulation that will be used to compare the different mesh elements in the mesh
independence study.

Fig. 2. Computational Model for RANS Simulation.

2.3 Turbulence Model

In a RANS model, the flow calculation is obtained thanks to the mean transport equations
of mass Eq. (1) and momentum Eq. (2). If the density of the flow is constant the Reynolds
mean (noted with the bar “ ”) is used in these equations. However, if the density is
124 G. Guevara-Morales et al.

not constant, (which is the case in reactive flows with great temperature gradients) the
Favre mean (noted with the tilde “∼”) is adapted to resolve the equations. The equations
obtained with this mean are analog to the equations obtained with the Reynolds mean
and the mean density of the products is taken into account. In Eqs. 1 and 2, variables m S
and FS represent source terms for mass and momentum respectively, this terms translate
the vapor and momentum transferred from the fuel drops to the gaseous mix.

∂ ρ̄ ∂ ρ̄ ũ j
+ = mS (1)
∂t ∂x j
∂ ∂ ∂ p̄ ∂τt j ∂
ut ) +
(ρ̄ (ρ̄
ut
u j) = − + − (ρ̄ u
 u ) + F
t J S (2)
∂t ∂x j ∂ xt ∂x j ∂x j

The turbulent flow was modelled used a standard Low Reynolds κ-ε model. This
model was chosen because of its two-equation approach and versatility in this type of
problems and the low computational cost that it represents. Even though, in general, the
Reynolds Stress Model is more common in problems where recirculation is present, the
feeble swirl number inside the flow justifies the hypotheses for a standard κ-ε model.
The κ-ε model is based on Boussinesq’s hypothesis, which replaces the unknown
residual stress tensor (u u  ) that results from applying the Reynolds mean to the Navier-
t j
Stokes Eq. (2) equations with a “turbulent viscosity” (μt ) term that models the effects
of the residual tensor. The turbulent viscosity term is obtained via the turbulent kinetic
energy (κ) and the energy dissipation (ε) with the following expression, in which is
a model’s constant. The model resolves two different transport equations, one for the
turbulence’s kinetic energy Eq. (3), and one for energy dissipation, both equations present
the classical form of a transport equation, which is: Rate of change + Transport by
convection = Transport by diffusion + Production – Destruction. The production
and destruction rates, as well as the diffusive transport general coefficients are modelled
by empirical observation of various turbulent flows over the years and adjusted with the
different constants that make up the standard model, as shown in Eqs. 3 and 4.
In this study, the gaseous phase is modelled as mixture of ideal gases which density
only varies in function of temperature thanks to the fact that we consider the combustion
to happen at subsonic speeds and having therefore an isobaric flame. By using the ideal
gas law, the software is capable thereby capable of relating the density of the mix and
the temperature yielded by the simulation.
  
∂ ρ̄κ ∂ ρ̄ ut κ ∂ μt
+ = ∇κ + ρ̄ Pk + ρ̄ε (3)
∂t xi ∂x j σk
  
∂ ∂ ∂ μt ρ̄ Pkε ρ̄ε2
(ρ̄ε) + (ρ̄u iε) = ∇ε + C1ε − C2ε (4)
∂t ∂x j ∂x j σε k k

2.4 Spray Model

The injection and atomization of fuel are not directly simulated. The sub-model of a
pressure atomizer available in Fluent that was used on this paper is based on the LISA
 Numerical Simulation of the Formation, Evolution and Control 125

atomizing model (Linearized Instability Sheet Atomization) developed by Schmidt et al.

[13]. The LISA model determines the statistical mean of the spray’s drops using the
atomizer’s known parameters. The model is directly based on the classical atomization
of a fuel layer illustrated on “Atomization phenomena modelled by LISA (Fig. 3).
The model calculates, first hand, the dimension of the fuel layer (ts ) as a function
of mass, pressure, injection diameter, and spray angle. The parameters used for this
particular injector were a hollow cone type injector with a nominal angle of 60°. However,
when comparing the experimental results to the simulation this paper found that the drop
dispersion was better approximated by an angle of 58°. According to the NIST database
[14] the atomizer used on the experiment was a “Delavan WDA-1.0” with an outlet
diameter of 0.28 mm. The pressure and mass flow were reported by Widmann and
Presser [1] and are condensed, along the other injector operation parameters on Table 1.

Table 1. xx

Parameter Value
Air Temperature 300 K
Fuel Temperature 300 K
Fuel Flow 3 kg/hr
Injection Pressure 690 kPa
Injection Angle 60º
Operation Pressure 1 atm

After considering the aforementioned parameters, the model applies a perturbations

specter to the fuel layer with a depth of ts moving at the velocity Us through a static,
oxidant atmosphere, the most unstable perturbation, and consequently the one which
will break up the fuel film into drops, will be the one that reaches the highest growth rate
(Ω). Following this reasoning, the model will calculate Ω by studying the upper limits
of the growth rate equation. The equation translates the fact that growth rate is a function
of the wavelength “k”, therefore Ω will be yielded by the maximum wavelength “ks ”.
When the fuel layer perturbation with ks wavelength and growth rate “Ω”, reaches a
critical wave amplitude (ηb ), the fuel layer will break up into ligaments of “d L ” diameter.
This diameter is expressed in Eq. 5 which presents two possibilities: a) in the first case,
the wavelength is negligible compared with ts and therefore the diameter d L will depend
only of ks and the ligament constant C L or b) In this case, the wavelength is larger than
ts thus the ligament diameter d L will depend on ks and h f , which represents the width
of the fuel layer were the atomization occur. The distance (L b ) between this point and
the injection point is calculated thanks to the fuel layer’s velocity and the time elapsed
before atomization (τs ), i.e. L b = Us τs . Time (τs ) is given by the critical wave amplitude
equation Eq. (6). Considering the former, L b is expressed in Eq. 7 as a function of Us ,
Ω and the term ln(ηb /η0 ), also known as layer constant. The software affects a value of
12 to this constant thanks to the fuel’s feeble viscosity (when compared to water).
126 G. Guevara-Morales et al.

Fig. 3. Atomization phenomena modelled by LISA [15].

The value of 12 was corroborated by experimental results with different fuels which
Weber’s numbers varied between 2 and 200 [16]. The width (h f ) is computed thanks
to L b and the initial fuel layer width. The initial drop diameter (d0 ) of drops resulting
from the atomization of the ligaments with a d L diameter is determined thanks to Eq. 8.
Said equation is based on the wavelength of the wave caused by the breakup of the
ligament (λopt ), equating the volume of the ligament and the resulting spherical drop.
The value of d0 , yielded by the LISA model is equated to the mean diameter (d̄) in a
Rosin-Rammler drop distribution. In this function the mass fraction of drops that have
a diameter superior to a certain (d) diameter, is given by Eq. 8. Said equation yields Yd ,
the mass fraction and n, a parameter that translates the dispersion in drop size. For most
practical applications, n varies between 1.8 and 3 [17].
 
2πC L 8h f
dL = , (5)
ks ks
ηb = η0 eΩτs (6)
 
Us ηb
Lb = ln (7)
Ω η0
d0 = 1.88d L (1 + 3Oh)1/6 (8)

Given the value of d̄ and n, the injected drops diameter “d” is calculated by randomly
studying the mass fraction Yd in Eq. 9. For this paper’s purposes the value of n is not
calculated as it is a statistical value. The actual value of the parameter was studied in
a number of cases to determine the best match for the experimental database result,
after this empirical approach it was found that the best match for spray drop dispersion
is n = 4.5. Furthermore, the trajectory of injected drops is also random. However,
considering the spray angle of 58° and the dispersion angle presented by Zhu et al. [10]
it was concluded that the dispersion angle should be around 10°. After a certain number
of iterations, it was concluded that an angle of 15° has a greater degree of fidelity towards
 Numerical Simulation of the Formation, Evolution and Control 127

the atomization distances presented in the experimental results. The number of particles
injected into the domain was determined to be 8000 per time step, this enabled the mass
flow to be realistic and therefore went well to the rest of the spray mean parameters.
Once the drops are injected into the domain, thanks to the LISA sub-model, the
second atomization mode is simulated. The sub-model TAB (“Taylor Analogy Breakup)
that is used to simulate this second mode is based on Taylor’s analogy, which compares
a system composed of a spring and a mass and a drop’s deformation and oscillation. It
is given then that the drops will breakup into smaller drops as soon as their deformation
rate reaches a critical value. To calculate the drops’ trajectory the simulation takes into
account the mean values of flow velocity (ũ), we need however to consider the whole
velocity when determining trajectories, which creates a need to consider the random
fluctuations of the mean velocity value throughout time, in order to do this the simulation
used the “random walk” probabilistic model which introduces an eddy whose proper
velocity is noted u , the eddies are introduced throughout the particle´s trajectory. By
using this model, the velocity considered before integrating the position equation of the
drop is: ũ + u  (t) The value for u  (t) is calculated using a random probability variable.
Furthermore, collisions between particles are also taken into account in the spray model
and are accounted for with O’Rourke’s probabilistic algorithm. The algorithm considers
that two drops may only collide if both are inside of the same mesh element. To determine
the probability of a collision happening a collision volume is computed. The probability
of collision is then computed by calculating the ratio between the collision volume and
the mesh element volume. Once the algorithm has determined a collision, the results
of this collision were reduced to only two outcomes: either coalescence or an elastic
collision. The outcome of collision is defined by the relative position of both drops
during the collision and a random variable. The drop’s properties after collision are
calculated thanks to conservation laws.

2.5 Combustion Model

The turbulent diffusion flame is modelled through a steady non-adiabatic flame model
also known as flamelets, this is available in the non-premixed combustion models in
Fluent. This model was designed to simulate turbulent non-premixed flames. Using the
fundamental properties of a non-premixed flame whose reaction zone can be assimilated
to a surface, it is inferred that the flame divides the domain into two zones, a zone with
oxidant and combustion products outside the flame and a zone with fuel and combustion
products, inside the flame. Therefore, we can add the transport equations for mass fraction
and enthalpy. Besides, the hypothesis of a Lewis number equal to 1 throughout the flow
lets us assume that the diffusion coefficients for chemical species and for heat are equal.
To avoid having to calculate the chemical reaction rate and thanks to the previously
made hypothesis we use the Shvab-Zeldovitch formalism to write the transport equations
in terms of Zeldovtich’s variables. We norm these variables by using the mixture fraction
f defined using Zeldovitch’s variables. The software then solves for a transport equation
of f which is only a lineal combination of a Zeldovitch variables and therefore satisfies
the same transport equations. Applying Favre’s mean to the transport equation yields
Eq. 9. It is important to note that the production term for mixture fraction comes solely
128 G. Guevara-Morales et al.

from the evaporation rate of the drops within the domain, same as mass and momen-
tum in previous equations. Furthermore, the diffusion coefficient for mixture fraction is
considered to be defined by turbulence which has a greater effect on chemical species
diffusion than molecular diffusion by itself. Therefore, we consider the molecular diffu-
sion coefficient to be negligible in favor of a turbulent diffusion coefficient defined by the
“turbulent viscosity” that models Reynolds Stress Tensor. Besides a transport equation
for the variance of fraction mixture ( f 2 ), given in Eq. 10. Since the simulation takes
consideration of the steady flow the variation of mixture fraction and its variance (first
term for Eqs. 9 and 10) is zero. We define scalar dissipation (χ ) which characterizes
the narrowing of the flame from the mixture fraction gradient, i.e. χ = 2D|∇ f |2 . To
represent flames with different narrowing degree we can express scalar dissipation in
the stoichiometric zone of the mixture (χst ). The Favre’s mean χst is then assimilated
through a linear regression of the transport equation of χ [3], yielding Eq. 11.
  
∂ ∂ ∂ μt
(ρ̄ f˜) + (ρ̄ ũi f˜) = ∇ f˜ + f s (9)
∂t ∂x j ∂x j σf
  
∂ ∂ ∂ μt 2 ε 2
2
(ρ̄ f ) +  2
(ρ̄ ũi f ) = ∇ f 2
+ Cgμt ∇ f˜ −ρ̄Cd  f (10)
∂t ∂x j ∂x j σf k
ε
χst = Cχ f 2 (11)
κ
Before the simulation, the model generates a number of tables (one for each flame
with a certain narrowing degree) that resolve the steady transport equations (a flamelets
model considers the turbulent flame to be a superposition of laminar, steady flames that
vary in position and form over a defined step of time) and yield results for temperature and
mass fraction of each of the chemical species involved in the combustion problem, these
results are given in function of the mean values of f, χst and enthalpy (H ), even though
the latter does not affect the chemical reaction directly. The relation between temperature
T and mass fraction of the chemical species n(Yn ) in function of the variables f , χst y
H is yielded by Eqs. 12 and 13. This equations are obtained via the transport equations
of chemical species and enthalpy for a laminar flame, in which the coordinate system
can be assimilated to a 1D problem in which f varies (Shvab-Zeldovitch formalism),
therefore the convection transport is negligible in favor of the other terms.

∂Yn 1 ∂ 2 Yn
ρ = ρχ + ωn (12)
∂t 2 ∂f 2
 
∂T 1 ∂2T 1 1 ∂C p ∂Y n ∂ T
ρ = ρX − H nωn + ρX + C p, n (13)
∂t 2 ∂ f 2 Cp n
2C p ∂f n
∂f ∂f

The thermochemical information needed to solve for these equations is obtained

by the reaction mechanism of the fuel, in this case, methanol. The reaction mechanism
used in the simulation includes 167 elementary chemical reactions between 32 chemical
species. This kinetic mechanism has been developed and approved by Lindstedt et al. [18]
and is consultable in [19]. Using Eqs. 12 and 13 the software will calculate the weighted
average between the combustion variables and the probability density function of f .
 Numerical Simulation of the Formation, Evolution and Control 129

The function of the probability p( f ) is calculated using a beta PDF, which means it’ll
be calculated using the values of f˜ and f 2 . In the case of p(χst ) a delta function is
considered, meaning that the variance won´t be necessary for χst . This way a relation
exists between combustion variables and f˜, f 2 χst and H̃ (The latter are the variables
yielded by the simulation).
In the same way, as previous simulations, flame-emitted radiation will be taken into
account by the Discrete Ordinates model (“DO”). This models discretizes the whole
range of solid angles that are necessary to calculate the radiation in the heat transfer
term of energy transport. In contrast with other models such as Rosseland’s, DO deals
with the whole optical depth specter, this makes it adequate to simulate the radiation
coming from combustion.

3 Presentation and Discussion of Results

3.1 Comparison with Experimental Results

First, the flow velocity profiles are compared with the experimental results (reported at
9.5 and 17.6 mm in height with respect to the injection point). Figures 4, 5 and 6 show
the axial, radial and tangential velocity profiles respectively. The discrepancy between
the velocity values of the three components and the experimental results in the range of
approximately 0 to 15 mm of the radius is notorious. A similar difference is reported by
Zhu et al. [17] and Collazo et al. [2]. This discrepancy is due to the increase in speed
produced by the spray on the gaseous medium. The software determines this increase
by calculating the sum of momentum lost by the drops that cross an element of the
mesh, later this moment is introduced as a source term to the equations of the flow
of the corresponding element. This method takes the point effect of the particles and
distributes it “on the mesh element, overestimating the amount of fluid that is affected by
the drops. On the contrary, in the PIV measurement the measured velocity corresponds
to the fluid found between the drops of fuel, and that does not reach to be affected by its
displacement, so that in the experiment the punctual effect of the drops is not relevant,
while in the simulation it increases. As can be seen in “Axial velocity at 9.5 and 17.6 mm
of the injector (Fig. 4)” the difference in the area near the axis of the injection is greater
in the axial velocity profile, a component that is mostly affected by the speed of the
spray, while the tangential velocity profile (Fig. 6) has the smallest difference in this
area.
After this discrepancy interval, the results obtained from the simulation consistently
reproduce the evolution of the flow. The increase in radial velocity in the axial direction
(Fig. 5), due to the swirling expansion in the chamber, is adequately reproduced by
simulation. On the other hand, the increase in the maximum axial speed in the simu-
lated profiles (Fig. 4) when increasing the axial distance with the injector is due to the
increase in temperature. Considering the uncertainty of the experimental data, including
the uncertainty of the 1.4 mm velocity profile of the injector based on which the input
profile was defined, the velocity results given by the simulation are acceptable.
The results of the number of drops density (number of drops per cubic centimeter)
along the radial axis and in different axial planes are compared in “Number of drops
130 G. Guevara-Morales et al.

density (drop/cm3) at 5, 25, 45 and 65 mm (Fig. 7)” and “Sauter diameter at 5, 25, 45
and 65 mm (Fig. 8)”. As noted, the dispersion of the drops increases along the axial
axis, as the spray expands. Although the increase in the dispersion of the drops and their
displacement along the radius is adequately captured by the simulation, the discrepancy
in the peaks of maximum concentration is notorious. This suggests, in addition to the
lack of calculated dispersion of the drops, that the software considers more drops, which
means that a smaller average diameter is calculated. Compared with the results of the
number of drops density reported by Zhu et al. [2], the results of the present work show
greater similarity with the experiment.
The average diameter of Sauter is an important characteristic of the spray when the
interface between the drops and the medium is relevant, as in the case of combustion.
Reproducing this characteristic properly is necessary to make the spray behavior similar
to the experiment. In Fig. 9 and 10 it is observed that in general the total surface of the
drops is overestimated in the simulation, which suggests that the diameter of the drops
calculated in general is smaller than the experimental one. This would imply that the
number of drops is greater, since the same volume of fuel must be distributed in smaller
drops, which is consistent with the increase in the number of drops density. However,
the diameter trend is adequately reproduced by simulation. The slope observed in the
distribution of the Sauter diameter can be explained considering that the diameter of the
drops near the flame is smaller (due to the increase in evaporation) which translates into
a smaller volume/area ratio. On the other hand, the increase of the diameter with the
axial distance is possibly due to the fact that the drops of smaller diameter and greater
surface evaporate with greater speed, so that the drops of greater diameter remain at a
greater distance from the injector.
The “Temperature contours (K) (Fig. 11)” shows the temperature profile inside the
combustion chamber, where the candle-shaped flame is observed, which is congruent
with the conical shape of the spray and similar to the results of the previous simulations
[2–4]. The maximum temperature reached, according to the simulation, is approximately
1790 K. No experimental data on the temperature inside the chamber is reported, however

Fig. 4. Axial velocity at 9.5 and 17.6 mm of the injector

 Numerical Simulation of the Formation, Evolution and Control 131

Fig. 5. Radial velocity at 9.5 and 17.6 mm of the injector

Fig. 6. Tangential velocity at 9.5 and 17.6 mm of the injector

the result is similar to the previous simulations, where Zhu et al. [2] report a maximum
temperature of 1850 K, Collazo et al. [3] of 2100 K and Crocker et al. [4] of 1900 K.
The temperature at the exit of the chamber is approximately 750 K. Since the output
of the chamber was modified it is not possible to directly compare this result with the
experiment. However, and as indicated by Zhu et al. [2], considering that the average
output temperature is 550 K according to the experiment, the output being located at a
point farther than in the simulation, the temperature of 750 K given by the simulation is
congruent.
132 G. Guevara-Morales et al.

Fig. 7. Number of drops density (drop/cm3 ) at 5, 25, 45 and 65 mm

Fig. 8. Sauter diameter at 5, 25, 45 and 65 mm

 Numerical Simulation of the Formation, Evolution and Control 133

Fig. 9. Axial Velocity at 5, 25, 45 y 65 mm of the injector.

Fig. 10. Radial Velocity (15, 35 y 55 mm).

134 G. Guevara-Morales et al.

Fig. 11. Temperature contours (K).

The velocity vectors are plotted in “Flow velocity vectors (m/s) (Fig. 12)”, where
three zones with low intensity recirculation are observed. This recirculation is due to the
pressure gradient between the inlet air flow and the fuel flow. The fuel flow, which is faster
than the incoming air flow, produces a decrease in static pressure, which attracts part of

Fig. 12. Flow velocity vectors (m/s).

 Numerical Simulation of the Formation, Evolution and Control 135

Fig. 13. Spray droplet diameter (μm).

the inlet air flow. The small lateral recirculation zones are congruent with the axial veloc-
ity profile reported experimentally. On the other hand, the expected vortex rupture does
not occur, since an abrupt loss of the eddy structure or a larger recirculation zone is not
observed.
The “Spray droplet diameter (μm) (Fig. 13)” shows the distribution and properties
of the fuel particles (droplet packages) that are in the computational domain during the
last iteration of the simulation (the spray is transitory). This figure shows the distribution
of the size of drops, where it can be seen that inside the cone the drops are smaller (due
to the proximity with the flame). This is consistent with the Sauter diameter distribution
discussed above. All particles evaporate completely and incomplete trajectories are not
reported (particles that escape the domain or collide with its walls). Comparing the
distribution of the drops with the temperature profile, it is observed that some drops
cross the high temperature zone, which is congruent with the percentage of unburned
methanol experimentally reported (although in the simulation it is assumed that this
vapor also reacts). The maximum distance to the injection point reached by the spray is
approximately 297 mm.
Once the numerical technique to reproduce turbulent spray combustion phenom-
ena was valid, the injection process control was studied, setting the atomizer operating
parameters and varying the Reynolds number, the graphs of temperature contours of
Fig. 14, show us the different events if the injection velocity increases, so that the dis-
persion of the drops are affected during the combustion process in control of the size of
the drops decreases, the recirculation of the flow intensifies, the temperature decreases
until the flame goes out.
136 G. Guevara-Morales et al.

Fig. 14. Temperature Contours at different Reynolds

4 Conclusions

In the present work, the turbulent spray combustion process was simulated in the exper-
imental combustion chamber model developed by NIST. Experimental data have been
reported in other works, as well as the boundary conditions used in this simulation.
The combustion model used is the flamelet model, which demands few computational
resources considering that the reaction is rapid compared to the turbulent processes of
the flow. To simulate the spray, the LISA model was used, in addition collisions between
drops and their turbulent dispersion are considered. The data suggests that the droplet
size determined by the LISA model is smaller than the actual spray droplet size. The
results of the RANS simulation were used to estimate the turbulence scales present in
the flow.
The control of the injection of drops, the atomized or pulverized both are of the
utmost importance during combustion, since they depend on the quality of the burn,
the maximum temperature in the chamber, the number of oxidants and the amount of
contaminant products that are discarded to the atmosphere, so specialized studies in the
control of injectors are able to deliver adequate information to keep the flame on.

References
1. Widmann JF, Presser C (2002) A benchmark experimental database for multiphase combus-
tion model input and validation. Combust Flame 129:47–86
 Numerical Simulation of the Formation, Evolution and Control 137

2. Pope SB (2000) Turbulent flows. Cambridge University Press, Cambridge

3. Veynante D, Vervisch L (2002) Turbulent combustion modeling. Prog. Energy Combust. Sci.
28:193–266
4. Peters N (1986) Laminar flamelet concepts in turbulet combustion. In: Twenty-First
Symposium (International) on Combustion, pp. 1231–1250
5. Warnatz J, Maas U, Dibble RW (2006) Combustion, physical and chemical fundaments,
modeling and simulation, experiments, pollutant formation, 4 ed
6. Magnuseen BF (1989) Modeling of NOx and soot formation by the Eddy Dissipation Concept.
International Flame Research Foundation
7. Magnussen BF (1981) On the structure of turbulence and a generalized eddy dissipation
concept for chemical reaction in turbulent flow. In: 19th American Institute of Aeronautics
and Astronautics Aerospace Science Meeting
8. Zhou LX et al (2016) A review on studies of a SOM combustion model for single-and-two-
phase combustion. Int. J. Heat Mass Transf. 96:154–163
9. Zhou L, Li K, Wang F (2012) Advances in large-eddy simulation of two-phase combustion
(I) LES of spray combustion. Chin. J. Chem. Eng. 20(2):205–211
10. Zhu S, Roekaerts D, Meer Tvd (2011) Numerical study of a methanol spray flame
 Control of a Stewart-Gough Platform for
Earthquake Ground Motion Simulation

Erick Alvarado Requena(B) , Antonio Estrada , Gengis Toledo Ramı́rez ,

Noe Reyes Elias, Jorge Uribe, and Berenice Rodrı́guez

Centro de Ingenierı́a y Desarrollo Industrial, 76125 Querétaro, QRO, Mexico

e.alvarado@posgrado.cidesi.edu.mx

Abstract. The Stewart-Gough platform is a six DOF parallel robot

manipulator with higher precision, mechanical stiffness and load capac-
ity compared with a conventional serial link arm. In this work, a CAD
model of CIDESI Stewart Platform HxCf1 is imported to Simscape as a
dynamic model and an application of an inverse dynamics control scheme
is proposed; full kinematic equations of the robot are presented. A ground
motion movement “profile” consisting on position (x, y, z) and orienta-
tion (α, β, γ) based on real earthquake records is loaded and simulated;
simulation results illustrate the performance of the HxCf1 robot using a
standard PID controller.

Keywords: Parallel manipulator · Robotics · Control

1 Introduction
Parallel manipulators offer some advantages such as precision, mechanical stiff-
ness and higher load capacity, due to its parallel structure if compared with serial
manipulators; on the other side, a disadvantage of parallel manipulators is the
workspace limitations due to the link lengths and the mechanical limits of the
passive joints [1]. Generally, because of the non-linearity and the complexity of
the equations, forward kinematics of parallel manipulators is very complicated
and difficult to solve; this is also in contrast with serial manipulators. Dynamical
analysis of parallel robots is very important to develop a model based controller
but also complicated due to the existence of multiple closed-loop chains. Dynamic
equations of a Stewart-Gough platform can be derived based on Lagrange’s for-
mulation [2]; in this paper, application of an inverse dynamics control scheme is
proposed for trajectory tracking on a Stewart-Gough platform.
The CIDESI Hexapod robotic system is a project that aims to develop an
industrial grade robotic system that provides functionalities similar to those
usually found in an industrial robot; the CIDESI Hexapod has been designed
to be operated by a human operator (User), or by another system or device
(Application) e.g., a vision system.
Generally, a Hexapod robot system consists of a Robot (mechanical struc-
ture), controller, Teach Pendant, computer, user documentation and accessories.
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 138–146, 2020.
https://doi.org/10.1007/978-3-030-45402-9_14
 Control of a Stewart-Gough Platform for Ground Motion Simulation 139

These are the basic components that are generally included in any serious com-
mercial robotic system. The “Hexapod CIDESI project” is divided at least in
three phases, with the development of one prototype with increased capabilities
in each phase; the current prototype corresponds to the phase I, which is called
“Hexapod CIDESI phase 1”, better known by its nickname: “HxCf1” [3,4]. In the
Phase I, the robot is mainly intended as a low cost industrial-like experimental
prototype [5].

2 Stewart-Gough Platform Model

The Stewart - Gough platform is a robot designed in a closed chain, in which
a fixed platform in space, considered as the base of the robot, and a mobile
platform are connected; both joined by six linear extensible legs. The points of
union with the platforms are mechanical joints with more than two degrees of
movement; this manipulator is actuated by motors located on the prismatic joints
[6]. The design provides high performance in precision and strength compared
to serial manipulators.
In Fig. 1, the HxCf1 [7] prototype is shown. For modeling purposes two coor-
dinate frames are defined, a coordinate frame, B(x, y, z), attached to the fixed
base containing the universal joints Ai and a frame T(x, y, z) attached to the
mobile platform containing the universal joints GTi being i = 1 to 6. Extension
of each leg is denoted by li [8].

Fig. 1. CIDESI Hexapod HxCf1.

Hexapod positioners are based on 6 DOF motions; these are the three linear
movements x, y, z (lateral, longitudinal and vertical), and the three rotations
roll (x-axis), pitch (y-axis) and yaw (z-axis) of frame B. The following position
vectors of the connection points on its corresponding coordinate system are
derived from Fig. 1.
⎡ ⎤ ⎡ ⎤
rp cos(λi ) rb cos(υi )
GTi = ⎣rp sin(λi )⎦ ; Bi = ⎣rb sin(υi )⎦ (1)
0 0
140 E. Alvarado Requena et al.

The coordinate values for each point calculated using Eq. 1 are shown in
Table 1.

Table 1. Point coordinates for Bi and GTi .

Point Coordinates Point Coordinates

B1 (160.1226, 189.3407, 0) GT1 (189.2780, 36.9977, 0)
B2 (83.9125, 189.3407, 0) GT2 (−62.5981, 182.4184, 0)
B3 (−244.0351, 43.9999, 0) GT3 (−126.6799, 145.4207, 0)
B4 (−244.0351, −43.9999, 0) GT4 (−126.6799, −145.4207, 0)
B5 (83.9125, −233.3407, 0) GT5 (−62.5981, −182.4184, 0)
B6 (160.1226, −189.3407, 0) GT6 (189.2780, −36.9977, 0)

We refer to the origin of the moving platform as G. The position vector of G

w.r.t. the BXY Z coordinate system is denoted as P = [Px Py Pz ]T . The rotation
matrix of the moving platform relative to the fixed base is given by:
B
RT = RZ (γ)RY (β)RX (α) =

Thus, the vector Li corresponding to the link i is obtained as follows (in

BXY Z coordinates).

Li = RXY Z GTi + P − Bi , i = 1, ..., 6 (2)

and their lengths, li , correspond to the actuator positions

 
li = Li 

After obtaining inverse kinematics, a dynamic model of the Stewart-Gough

platform may be derived using the Lagrange formulation [2]. Kinetic and poten-
tial energies of the moving platform and the legs, each one consisting of a piston
and a cylinder, are obtained in compact expressions. Lagrangian formulations
are taken in compact symbolic forms in order to make the expression under-
standable. The matrix equation is showed below:

M  (Xp )Ẍp + Vm (Xp , Ẋp )Ẋp + G (Xp ) = J T (Xp) τ, (3)

Where M  (.) is the mass matrix, Vm (.) is the velocity product term, G (.) is
the gravitational term, J(.) is the Jacobian matrix and τ ∈ R6 is the generalized
actuator force at each actuator all with respect to the based frame.
Dynamic formulation of the hexapod in the joint space coordinates can be
described by transforming the dynamics written in the task-space coordinates
starting with the Jacobian Matrix J(.):

ẋ = J −1 q̇, (4)
 Control of a Stewart-Gough Platform for Ground Motion Simulation 141

where q̇ ∈ R6 is the velocity vector for the six legs. A dynamic matrix equa-
tion in the joint space coordinates will be obtained as follows:

M (q)q̈ + Vm (q, q̇)q̇ + G(q) = τ, (5)

where
M (q) = J −T M  (Xp )J −1 ,
Vm (q, q̇) = J −T M  (Xp )J −1 JJ
˙ −1 + J −T Vm (Xp , Ẋp )J −1 , (6)
−T 
G(q) = J G (Xp ),
Where q ∈ R represents the six link displacement vector.
6

3 Control Scheme Proposal

In order to linearize and decouple the robot manipulator dynamics, inverse
dynamic control is used; possible non-linearities such as Coriolis and centrifugal
terms as well as gravity terms can be compensated by adding these terms to the
control input. However, this type of feedback linearizing controller is much more
complicated and it’s conception is not straightforward for the robot manipulator
dynamics [9].
In order to cancel non-linear terms and decouple the dynamics of each link,
it is used an inverse dynamics control of the form:

u = M (q)u0 + C(q, q̇)q̇ + G(q), (7)

applied to (3). The mass matrix, M (q), is a symmetric positive definite matrix
which verifies:
λm I ≤ M (q) ≤ λM I, (8)
where λm (λM < ∞) denotes the strictly positive minimum (maximum) eigen-
value of M (q) for all configuration q. These properties show the regularity of the
inertia matrix. The controller of (7) and the regularity property of M (q), yield
a set of n decoupled linear systems.

q̈ = u0 (9)

where u0 is an auxiliary control input to be designed. A typical choice for u0

is:
u0 = q̈d + KD (q̇d − q̇) + KP (qd − q), (10)
which leads to the error equation

q̃¨ + KD q̃˙ + KP q̃ = 0, (11)

where q̃ = qd − q = −e.
The error equation, (11), is exponentially stable by a suitable choice of the
matrices KD and KP ; Eq. 11 can be rewrite in its state space form, ξ˙ = Aξ,
where the matrix A is Hurwitz provided KD and KP were suitably chosen.
142 E. Alvarado Requena et al.

4 Dynamic Model
The dynamic model of the CIDESI HxCf1 Hexapod was made taking as base
the CAD model in Simscape software. Simscape enables to create models of
physical systems within the Simulink environment; it is possible to build physical
component models based on physical connections that directly integrate with
block diagrams and other modeling paradigms.
Simscape helps to develop control systems and test system-level performance.
The block diagram modeling environment for the engineering design and sim-
ulation of rigid body machine and their motions, allows the user to model and
simulate mechanical systems with a suite of tools to specify bodies and their
mass properties, their possible motions, kinematics constraints and coordinate
systems, and to initiate and measure body motions. The full model with its
components is shown in Fig. 2.

Fig. 2. CIDESI HxCf1 Hexapod Simscape model. 1. Kinematic blocks, 2. Jacobian

blocks, 3. Configurable PID controller, 4. Assembly exported from SolidWorks, 5. Posi-
tion sensor of the mobile platform, 6. Leg position and error, 7. Mobile platform position
error.

The exported CAD model from SolidWorks to Simscape using SimMechanics

add-on generate one XML file and a set of geometry files that were converted in
order to create the platform part by part by linked blocks. After the conversion
was complete, the geometry blocks of the platform looks as depicted in Fig. 3
and the visual output of the platform is shown in Fig. 4.

Fig. 3. CIDESI HxCf1 Hexapod exported geometry.

 Control of a Stewart-Gough Platform for Ground Motion Simulation 143

Fig. 4. CIDESI HxCf1 Hexapod visual output.

5 Simulation
A simulation study will be performed to illustrate the defined Simscape model
of the HxCf1 Hexapod, using an earthquake ground motion “profile”, see Fig. 2,
consisting on 3 positions (x, y, z) and 3 orientations (α, β, γ) based on real ground
motions that was obtained from PEER database [10] and loaded as input. In the
simulation, the reference positions are updated at a 50 ms rate and a classic PID
controller is used, setting the gains as follows: Kp = 100, Ki = 20 and Kd = 2.
In order to perform a simulation, an input file containing the required posi-
tions and orientations for the Stewart Platform is used; for this simulation a
real earthquake ground motion obtained from PEER database [10] was used as
input. In the simulation, the reference positions are updated at a 50 ms rate
and a classic PID is used; after running the simulation, the values given by the
position sensor in the mobile platform are compared with the desired values and
the length errors of the actuators 1, 3 and 5. The results are shown in Fig. 5. For
the mobile platform, the position sensor data was plotted against the desired
position data in the x, y, z axis; plots are shown in Fig. 6 having the error plots
in Fig. 7.

Fig. 5. Actuator length errors

144 E. Alvarado Requena et al.

Fig. 6. Mobile platform axis position.

Fig. 7. Mobile platform axis error.

6 Experimental Results

In order to perform experimental tests and be able to compare the results with
the Simulink dynamic model, an Arduino program using FreeRTOS library was
written; the program allow to read the measurement of the position of the actu-
ators, using their own potentiometer, at a specified time.
With the measurements done, the real position of the actuators can be com-
pared with the desired position that was used as input to the Stewart Platform
at different points of the routine. The plots shown in Fig. 8 display the perfor-
mance of the experimental readings against the desired positions in the x, y, z
axis for the actuators numbered 1, 3 and 5; the plots shown from Fig. 9 show
the error in the same referred axis.
The measured response of the actuator potentiometers is very noisy for each
measurement. It was necessary to use a filter in order to obtain a more realistic
result. After passing the data trough a first order filter with a time constant of
1 s, the corresponding plots are shown in Fig. 10.
Control of a Stewart-Gough Platform for Ground Motion Simulation 145

Fig. 8. Measured positions against desired positions.

Fig. 9. Mobile platform axis position.

Fig. 10. Filtered experimental error.

146 E. Alvarado Requena et al.

7 Conclusions
The simulation of the earthquake profile was very fast and the resulting error is
very low in every axis. For the experimental tests, it is evident that the measured
position values at each point are following the reference values. Nevertheless,
the error is quite larger than the results in simulation. Additionally, there is
the problem of the noise in measurements of positions, this could be due to
parasitic current that are returning to the voltage source when the routine is
being performed; the filter that was applied help to obtain results closer to
reality since the spikes on measured positions do not represent a real feasible
motion of the platform. It is important that for further work to eliminate most
of the noise generated in the measurements in order to be able to experiment
with more earthquake “profiles” with different frequencies and have a better
evaluation of the HxCf1 performance.

References
1. Merlet J-R (1995) Determination of the orientation workspace of parallel manipu-
lators. J Intell Rob Syst 13:143–160
2. Ghobakhloo A, Eghtesad M (2006) Closed form solution to inverse dynamics prob-
lem of a parallel robot (Stewart platform) by Lagrangian approach. In: Tehran
international congress on manufacturing engineering (TICME 2005)
3. Toledo-Ramı́rez GK (2018) Hexápodo-CIDESI, fase 1. conceptos de operación y
requerimientos. Technical report, CIDESI-CONACYT, August 2018
4. López GAD (2018) Desarrollo de prototipo alfa de plataforma Stewart-Gough
“Hexápodo CIDESI”. Technical report, CIDESI-CONACYT
5. Germán Félix JL (2018) Robot controller for a parallel robot, from conception to
implementation. Master’s thesis, CIDESI-CONACYT & FH Aachen
6. Stewart D (1965) A platform with six degrees of freedom. In: Proceedings of the
institution of mechanical engineers
7. Barasja Fletes JR, Toledo Ramı́rez GK (2018) Diseño de detalle de prototipo alfa
Hexápodo CIDESI - Fase I. Parte Mecánica. Technical report, CIDESI-CONACYT
8. Arciniega E (2018) Cinematic and dynamic simulations of CIDESI Hexapod HxCf1.
Technical report, CIDESI
9. Ghobakhloo A, Eghtesad M (2006) Position control of a Stewart-Gough platform
using inverse dynamics method with full dynamics. In: Advanced motion control
10. Users manual for the PEER ground motion database web application. https://
ngawest2.berkeley.edu/peer ground motion database users manual.pdf. Accessed
19 Apr 2019
 Flow Control on a Wing Section Through
the Implementation of the SDBD Plasma
Actuator

R. A. Bernal-Orozco(B) , A. Arias-Montano, and O. Huerta-Chávez

IPN-ESIME Unidad Ticoman, SEPI. Av. Ticoman 600, San José Ticoman,
C.P. 07340 Gustavo A. Madero, Ciudad de México, Mexico
rulozco@hotmail.com, aarias@ipn.mx, lecram 21@hotmail.com

Abstract. This work presents a Computational Fluid Dynamics (CFD)

wind flow analysis technique applied to a Wing Section using a two-
dimensional Reynolds-Averaged Navier-Stokes Method. The behavior of
the flow has been studied from different angles of attack of the aerody-
namic section until the angle of stall is found. An active flow control
by means of the SDBD plasma actuator is implemented numerically,
through body force source terms in the Navier-Stokes equations to re-
attach the boundary layer of the flow. Then, altering the position and
frequency of the SDBD plasma actuator results in the delay of the stall
beyond 16◦ and an increase of lift with reduction of the drag for new
critical angles of stall.

Keywords: Flow-control · SDBD · Dielectric barrier discharge ·

Airfoil with plasma actuator · RANS

1 Introduction

The lift of an airfoil depends greatly on the attachment of the flow. One way
to keep the flow attached is through the use of flow control methods such as
Single Dielectric Barrier Discharge (SDBD) actuators. A flow control method as
defined by Flatt [4] is “any mechanism or process, which alters the boundary
layer so it behaves different from it will naturally do”. An SDBD actuator is a
device that through an electrical discharge ionizes the air and by the acceleration
of the ions by the electromagnetic forces, produces a transfer of momentum to
the flow, thus diminishing its separation, Fig. 1 shows a schematic diagram of a
plasma actuator.
The use of SDBD actuators in aerodynamic applications such as reattach-
ment of flow in laminar separation bubbles was proved by Corke [2] who suc-
cessfully reattached the flow in a NACA 663 − 018 airfoil up to 8◦ beyond the
stall. Corke also tested them in blades from a low pressure turbine of a gas

c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 147–158, 2020.
https://doi.org/10.1007/978-3-030-45402-9_15
148 R. A. Bernal-Orozco et al.

turbine engine. Greenblatt [6] proposed its use in wind turbines, with the use of
an actuator he obtained a 7% increase in the net power delivered by the wind
turbine.

Fig. 1. SDBD scheme.

In the present study, two-dimensional Reynolds-averaged Navier–Stokes

equations (RANS) simulations of an airfoil with an SDBD at a Reynolds number
(Re) of 2.0 × 105 are conducted, to analyze the effect of changing the position
and frequency of the actuator. While most previous studies used commercial
software, in this work the simulations are performed with the open source soft-
ware OpenFOAM. The actuator was implemented as a source term body force
so it can be directly introduced in the flow equations, as explained in Sect. 3.
The position of the actuator is analyzed by placing the actuator at 3% and then
at 10% of the airfoil chord for several angles of attack, as shown in Sect. 6.1. For
the frequency analysis the base frequency remain fixed and the burst frequency
was changed in a range from 3 Hz to 45 Hz, this is presented in Sect. 6.2. In Sect.
7 the methodology used in this study was validated through comparison with
experimental data.

2 Flow Equations

To simulate the flow around the airfoil. The RANS equations are used. Taking the
flow as incompressible and by the fact that most of the electric energy consumed
by the actuator is transformed into kinetic energy, the change in temperature
can be neglected, so it is only required the continuity equation, Eq. (1), and the
momentum equation Eq. (2).
∂ u¯j
=0 (1)
∂xj
 
∂ ūi ∂ ūi 1 ∂ p̄ ∂ ∂ ūi fb
+ u¯j =− + (ν + νt ) + (2)
∂t ∂xj ρ ∂xj ∂xj ∂xj ρ
where p is the pressure, u the speed, ρ the density, ν the kinematic viscosity, νt
the eddy viscosity, and fb the body force of the actuator.
From the turbulence models available in OpenFOAM the k − ωSST was
selected. It is a lineal model, in which the Reynolds stress are modeled by the
 Flow Control on a Wing Section 149

Boussinesq hypothesis which is a linear relationship: −ρui uj  = 2μt Si,j − 32 ρkδij ,

where μt is the eddy viscosity, k the turbulence kinetic energy and Si,j the strain
rate tensor. Then the transport of energy from the turbulence fluctuations is
modeled through the eddy viscosity.
The k − ωSST model is characterized by being a two equations model, one
for turbulent kinetic energy Eq. (3), and another for specific dissipation Eq. (4).
This model uses a k − ω formulation in the inner part of the boundary layer
from the wall up to the viscous sub-layer. This formulation changes the behavior
to k −  model in the free stream, thus avoiding the problem of over sensitivity
of the k −  model to the turbulence intensity in the free stream. It is a model
capable of capturing separation.
∂k ∂k ∂ui
+ Uj = τij −
∂t ∂xj ∂xj
 
∗ ∂ ∂k
β ωk + (ν + σk1 νt ) (3)
∂xj ∂xj

 
∂ω ∂ω γ1 ∂ui ∂ ∂k
+ Uj = τij − β ∗ ω2 + (ν + σω νt )
∂t ∂xj νt ∂xj ∂xj ∂xj
1 ∂k ∂ω
+2 (1 − F1 ) σω2 (4)
ω ∂txi ∂xi
The turbulent viscosity νt in the model k − ωSST is defined by the Eq. (5).
a1 k
νt = (5)
max (a1 ω, b1 F23 S)

3 Modeling of the Actuator

To simulate an SDBD plasma actuator its computational model is required.
Various models have been developed, those seek to accurately reproduce the
properties of plasma, such as ion formation rate, collisions. But them entails
an increase in the computational cost making them impractical for engineering
applications. Alternatively engineering models such as Suzen model [11] or Shyy
model [10] have been developed, these models are based on the idea that the body
force is directly proportional to the product of the charge density by the electric
field. However, these models have some drawbacks such as that the electrodes
should be strictly modeled in the domain since the thickness and position are
required by the equations that determine the electric field. Besides that in the
solver the electrical equations must be included.
That is why in works such as those carried out by Dörr and Kloker [3] the
body forces are modeled as source terms. Kloker states that a SDBD can be
modeled as a body force parallel to the wall. In the work performed by Sato [9]
he used the Suzen model to calculate the magnitude of the force that is applied
as a source term directly in the momentum equation.
150 R. A. Bernal-Orozco et al.

In this work an actuator with a length of 0.03x/c was modeled, by solving

the Eq. (6) from the Kloker model, the extension of the body was obtained, the
body force extends from the wall up to a height of 0.003x/c, in the Fig. 2 the
extension of body force is shown.
  
f (x, y) = cx a0 a1 x + a20 a2 x2 e−a0 x
  2/5
· b1 y + b2 y 2 e−b0 y (6)

Fig. 2. Extension of the body force for an actuator 0.03x/c length.

The magnitude of the body force was selected from the results obtained
by Hofkens [7] in which he determined the strength of a plasma actuator from
velocity field data. The results obtained by Hoefkens are shown in Table 1.

Table 1. Body force at different voltages for fbase = 2 kHz.

Voltage kV Force N/m3

8 416
10 2038.8
12 2308.6
14 3258.8
16 3302.6

Plasma actuators can operate both in a continuous mode or through pulses

as in the burst mode. It has been proven that an actuator operating in burst
mode is more energy efficient [1,5,9,12].
The burst mode is generated multiplying a low frequency square signal with
a high frequency sinusoidal base signal. In Fig. 3 is shown a scheme of the burst
 Flow Control on a Wing Section 151

Fig. 3. Signal scheme for burst mode. Fig. 4. Duty cycle for Fburst = 15 Hz
and fbase = 2 kHz.

mode, and in Fig. 4 the graph generated by a burst frequency of 15 Hz and a

base frequency of 2 kHz is shown.
The burst mode depends on the burst ratio Eq. (7), the burst frequency
Eq. (8) and the base frequency Eq. (9).

Ton
BR = (7)
T
F + = Fburst L/U∞ (8)
f + = fbase L/U∞ (9)
where L is the length reference (airfoil chord) and U∞ is the free stream velocity.
A body force of 416 N/m3 was selected, which corresponds to a voltage of
8 kV and a base frequency of 2 kHz, the burst frequency will have values of 3 Hz,
15 Hz and 45 Hz.

4 Geometry and Meshing

A structured C-type mesh was generated with blockMesh, Fig. 5 and Fig. 6 show
the mesh. The airfoil has a unit chord of c = 1.0. Three meshes were used to
perform a mesh independence analysis, the properties of the meshes are shown
in the Table 2.
For the mesh study the values of the lift coefficient (Cl ) and drag coefficient
(Cd ) were compared for a range of angles of attack from 0◦ to 12◦ for a Re number
of 2.0 × 105 . Table 3 and Table 4 show the results of the mesh independence
study. Mesh 3 was selected to perform the simulations for the study of the
plasma actuator, the y + was kept under a value of 5, and the mesh has a length
of 25 × c and a height of 20 × c.
152 R. A. Bernal-Orozco et al.

Fig. 5. General aspect of the mesh. Fig. 6. Mesh detail.

Table 2. Mesh properties.

Parameter Mesh 1 Mesh 2 Mesh 3

Nodes 629,700 775,500 104,7340
Cells 313,650 386,400 522,000
Max aspect ratio 306.773 309.543 277.867
Max non-orthogonality 29.171 29.5681 29.6446
Max skewness 0.45114 0.45267 0.45037

Table 3. Cl values for the mesh independence study.

α Cl Mesh 1 Cl Mesh 2 Cl Mesh 3

0◦ 1.29E–07 –3.81E-08 –4.30E–07
◦
4 0.3717 0.3739 0.3739
8◦ 0.7189 0.7223 0.7273
◦
12 0.9751 0.9793 0.9859

Table 4. Cd values for the mesh independence study.

α Cd Mesh 1 Cd Mesh 2 Cd Mesh 3

0◦ 0.0156 0.0154 0.0145
◦
4 0.0088 0.0093 0.0096
8◦ 0.0787 0.0795 0.0804
◦
12 0.1711 0.1724 0.1742

5 OpenFoam Setting up

The actuator was implemented in OpenFOAM as a source term using the utility
CodedSource in which through C++ language several types of source terms can
 Flow Control on a Wing Section 153

be programmed. Then the source term is introduced into a set of specific cells
(cell zone). The selected cell zone has a length of 0.03 × c and height of 0.003 × c,
which corresponds to the height established with the Eq. (6).
For the boundary conditions, a condition of type fixedValue is applied to
the inlet to set the free stream velocity, a condition inletOutlet which pro-
vides a generic outlet when applying zero gradient for positive flows (outside the
domain). A boundary condition of type noSlip sets the velocity with a value of
zero at the wall. Also over the airfoil sruface wall functions are applied for turbu-
lent kinetic energy kLowReWallFunction, turbulent viscosity nutkWallFunction
and for the specific turbulent dissipation omegaWallFunction. To ensure a two-
dimensional flow, a boundary condition of type empty is applied to the bound-
aries perpendicular to the flow.
The solver pimpleFoam was used and second order numerical schemes were
established, for the time a backward scheme was applied, for the gradient, diver-
gence and laplacians Gauss linear schemes with limiters, the interpolation were
lineal. The time step was modified in each iteration automatically based on a
maximum Courant number.

6 Results
The simulations were performed at Re = 2.0 × 105 with the following initial
conditions U∞ = 3.0 m/s, ρ∞ = 1.0 kg/m3 and ν = 1.5 × −5 m2 /s. The
performance of the actuator position and frequency were studied through its
effect over the lift and drag.

6.1 Actuator Position Analysis

Simulations of a NACA 0015 profile were performed at a Re of 2.0 × 105 for a
range of angles of attack from 0◦ to 16◦ to study the effect of the actuator along
of the chord, the actuator was placed at 3% (x/c = 0.03) and 10% (x/c = 0.1)
of the chord. In this case the actuator operated in its continuous mode.
Figure 7 shows the velocity field without actuator for an angle of attack
α = 16◦ , at this angle the airfoil is stalled. When the actuator is active the stall
is avoided and delays the detachment up to x/c = 0.5, this can be seen observed
in Fig. 8.

Fig. 7. Velocity field for α = 16◦ with- Fig. 8. Velocity field for α = 16◦ whit
out actuator. actuator at x/c = 0.03.
154 R. A. Bernal-Orozco et al.

The actuator reduces the detachment and prevent the stall, because the
electric discharge ionizes the air and due to potential difference between the
electrodes, an electromagnetic force moves those ions and that motion transfers
momentum within the boundary layer. The increase of momentum in the bound-
ary layer manifests itself as a jet within the boundary layer, see Fig. 9, that jet
modifies the velocity profile of the boundary layer to a positive velocity gradient.
The plasma actuator also increases the vorticity in the nearby region where
it is placed, this is evident in the graph of the Fig. 10 that shows two vorticity
peaks in the place where the actuator is located in each case x/c = 0.03 and
x/c = 0.1. The vorticity promotes the transition of the laminar boundary layer
to a turbulent boundary layer, in the turbulent boundary layer there is a greater
interaction between the flow outside it and that which is inside, so there is an
addition of momentum from the external flow to the boundary layer which helps
the flow to remain attached for longer.

Fig. 9. Boundary layer at x/c = 0.1 for Fig. 10. Vorticity (Ω) distribution for
α = 12◦ . α = 12◦ .

The pressure coefficient (Cp ) distribution is also affected, when there is a low
pressure peak near the leading edge (x/c = 0.0) as seen in Fig. 11 and Fig. 12, this
is more noticeable for the case at α = 16◦ . As consequence of the low pressure
increment the Cl increases.
The plot in Fig. 13 corresponds to the lift airfoil curve, in this graph is evident
that the actuator produce an increase of the Cl . Taking into account the effect of
the position of the actuator at different angles of attack, it is observed that the
actuator placed at x/c = 0.1 achieves the greater Cl for angles previous αmaxCl ,
and from α = 13◦ to α = 16◦ the actuator at x/c = 0.03 is more effective.
From the plot in Fig. 14 it is observed that there is an increase of Cd in
angles of attack smaller than α = 8◦ , within the range of α = 10◦ to α = 15◦
the actuator is able to reduce the drag by achieving a notable difference with
respect to the Cd that is present without an actuator, finally at the angle of stall
α = 16◦ due to the actuator effect the Cd decreases significantly because the flow
reattachment. It can be mentioned that there is no significant difference in the
 Flow Control on a Wing Section 155

Fig. 11. Cp distribution for α = 8◦ . Fig. 12. Cp distribution for α = 16◦ .

Fig. 13. Cl polar for Re = 2.0 × 105 . Fig. 14. Cd polar for Re = 2.0 × 105 .

effect of the actuator position over the Cd , having the same behavior regardless
of the actuator is at x/c = 0.03 or x/c = 0.1.

6.2 Actuator Frequency Analysis

For the analysis of the actuator frequency, 4 cases were studied for an angle of
attack of 16◦ , this angle of attack is of great interest because is when the airfoil
stalls, the cases are listed in the Table 5.
The plots in Fig. 15 compare the effect of the different burst frequencies over
the Cl for an actuation period of 1.5 s. It is observed that all the cases behave
like a step input, and with a burst frequency of 3 Hz, the highest Cl is achieved
additionally the increase is continuous and without any drops or over shoots.
The net effect of the actuator at different burst frequencies can be seen
through the lift and drag coefficients, in Fig. 16 the coefficients with and without
actuator are compared. In all cases with actuator there is an increase of the Cl
of more than 50% in comparison to the case without actuator, the maximum
Cl was achieved in continuous mode and with a burst frequency of 3 Hz. with
respect to Cd this remains virtually unchanged.
156 R. A. Bernal-Orozco et al.

Table 5. Cases for the study of the frequency of the actuator.

α◦ F + Fburst Hz f + fbase kHz

16 1 3 660 2
16 5 15 660 2
16 15 45 660 2
16 Continuous – 660 2

Fig. 15. Lift coefficient versus actuation time, for α = 16◦ .

In Fig. 17 the aerodynamic efficiency (Cl /Cd ) is compared, the highest aero-
dynamic efficiency was achieved with the actuator operating at a burst frequency
of 45 Hz, although it is evident that the variation in the efficiency of one case or
another is small.

Fig. 16. Aerodynamic coefficients vs Fig. 17. Aerodynamic efficiency vs

fburst , for α = 16◦ . fburst , for α = 16◦ .
 Flow Control on a Wing Section 157

7 Validation
To validate the methodology implemented in this work, simulations with and
without actuator were performed at a Re of 1.58 × 105 . The Re was used as
similarity parameter to compare the Cl from the simulations with the Cl reported
by Post [8], in which he used an SDBD in a NACA 0015 to attach the flow and
prevent the stall.

Fig. 18. Simulation and experimental Cl validation plot.

In Fig. 18 the Cl polar shows the results from Post and from the simula-
tions with the k − ωSST turbulence model. It is observed that the Cl behavior
from the simulations is similar to the one showed by the experimental data, so
there is a good correlation between the experimental data and the one from the
simulations.

8 Conclusions

The position of the actuator study showed that the actuator is more effective
when it is closer to the detachment point, when the separation occurs closer
to the leading edge the actuator placed at x/c = 0.03 performs better, if the
separation is closer to the trailing edge or x/c = 0.5 the actuator at x/c = 0.1
is better. If the detachment occurs before of the actuator it wont be able to
reattach the detached flow before it.
The graph in Fig. 13 shows that for angles of attack α ≥ 10◦ the Cl increase
is higher than for α ≤ 10◦ . For angles of α < 8◦ there is an increase of the Cd ,
at higher angles with greater separation regions the drag is reduced, as seen in
Fig. 14.
From the frequency study it was identified that the 3 Hz burst frequency
performs better as it achieves the highest Cl and the behavior of Cl during the
period of actuation shows a smooth and continuous increase. It was identified
that frequency variations do not have a noticeable impact on Cd . As in Fig. 16
158 R. A. Bernal-Orozco et al.

and Fig. 17 there is no significant magnitude variations over the Cl , Cd and β

regardless the frequency we suggest that a more detailed and sensitive frequency
study should be performed.
For future works it is intended to analyze the three-dimensional effects of
the actuator through Large Eddy Simulation (LES), and the implementation of
arrays with actuators placed at different positions of the airfoil in a way that
they can be controlled to be activated depending of the angle of attack in a way
that the actuator acts closer to the separation point.

References
1. Aono H, Sekimoto S, Sato M, Yakeno A, Nonomura T, Fujii K (2015) Computa-
tional and experimental analysis of flow structures induced by a plasma actuator
with burst modulations in quiescent air. Mech Eng J 2(4):15–00233
2. Corke T, Post M (2005): Overview of plasma flow control: concepts, optimization,
and applications. In: 43rd AIAA aerospace sciences meeting and exhibit, p 563
3. Dörr P, Kloker M (2015) Numerical investigation of plasma-actuator force-term
estimations from flow experiments. J Phys D Appl Phys 48(39):395203
4. Flatt J (1961) The history of boundary layer control research in the United States
of America. Bound Layer Flow Control 1:122–143
5. Fujii K (2014) High-performance computing-based exploration of flow control with
micro devices. Phil Trans R Soc A 372(2022):20130326
6. Greenblatt D, Müller-Vahl H, Lautman R, Ben-Harav A, Eshel B (2015) Dielectric
barrier discharge plasma flow control on a vertical axis wind turbine. In: Active
flow and combustion control 2014, pp 71–86. Springer, Cham
7. Hofkens A (2016) Determination of the body force generated by a plasma actuator
through numerical optimization
8. Post ML, Corke TC (2004) Separation control on high angle of attack airfoil using
plasma actuators. AIAA J 42(11):2117–2184
9. Sato M, Nonomura T, Okada K, Asada K, Aono H, Yakeno A, Abe Y, Fujii
K (2015) Mechanisms for laminar separated-flow control using dielectric-barrier-
discharge plasma actuator at low reynolds number. Phys Fluids 27(11):117101
10. Shyy W, Jayaraman B, Andersson A (2002) Modeling of glow discharge-induced
fluid dynamics. J Appl Phys 92(11):6434–6443
11. Suzen Y, Huang P, Jacob J, Ashpis D (2005) Numerical simulations of plasma
based flow control applications. In: AIAA paper, vol 4633, p 2005
12. West T, Hosder S (2012) Numerical investigation of plasma actuator configurations
for flow separation control at multiple angles of attack. In: 6th AIAA flow control
conference, p 3053
 Experimental Platform for Modeling
and Control of Articulated Bodies

Ana Hernandez-Maldonado1(B) , Juan G. Diaz-Martinez1 , Hirohisa Kojima2 ,

Irandi Gutierrez-Carmona1,2 , and Sajjad Keshtkar1,2
1 School of Engineering and Sciences, Tecnológico de Monterrey,
89600 Altamira, Tamaulipas, Mexico
A01410096@itesm.mx, juan.diaz@tec.mx
2 Department of Aerospace Engineering, Tokyo Metropolitan University, Hino, Tokyo
191–0065, Japan
hkojima@tmu.ac.jp

Abstract. Articulated bodies, like robotic manipulators and tethered satellites,

have gained great attention in recent decades due to their application in several
space solutions, as active debris removal, momentum transfer systems, electro-
dynamic reboots of space stations, among others. The dynamics of these mech-
anisms are usually considered as rigid pendulums or dumbbells. Several studies
have been carried out to validate the mathematical modeling and control either
of tether satellites with a massless tether and thus, no inertia; or rigid pendulums
under earth conditions. The aim of this paper is to combine the prior knowledge of
both proposals and present a mathematical model of the articulated space system
in orbital conditions. Numerical and experimental results, based on the ground-
based experimental setup, are considered to validate this model. This report also
includes technical information about the hardware and software components used
in the communication and connections of an experimental platform designed to
emulate different orbital trajectories.

Keywords: Kane formulation · Articulated space structures · Tethered

satellites · Orbital ground based experiment · Space robotics · Space pendulum

1 Introduction
Articulated satellite systems, consisting of two or more artificial satellites connected to
each other in space varies from traditional spacecraft with some aspects. One of the main
characteristics of these systems is a large length providing a stable orientation along the
local vertical and the creation of a small artificial gravity. Another important feature is
changeable configuration flexibility: the ability to change the length and orientation of the
connecting body allows the realization of different operations. This phenomenon allows
adjusting the relative position and orientation of spacecraft, attaching and detaching
objects from tethers, moving payloads along the connecting rod, and so on [1, 2].
One of the possible areas of application of these architectures is space debris removal.
Nowadays, pollution isn’t an exclusive problem of the earth, this phenomenon has
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 159–169, 2020.
https://doi.org/10.1007/978-3-030-45402-9_16
160 A. Hernandez-Maldonado et al.

reached dimensions from outer space, it’s estimated that there are more than 17,000
objects in LEO, including rocket bodies, spacecraft, and fragmentation debris, however
because of the high complexity of track small debris is impossible to know the real
amount for sure [3]. Many solutions have been proposed to this problem: Harpoons,
nets, tether satellites, and articulated bodies, such as robotic manipulators.
An articulated satellite system is a complicated physical system, therefore, most
research on dynamics and control of these systems use simplified models to make pre-
dictions about its behavior. These simplifications, however, may introduce a lack of
accuracy since important phenomena, as inertia, may not be considered. In many recent
studies, the TSS structure has been considered as a dumbbell model (Fig. 1). The pro-
posed model assumes that two satellites are connected by a rigid tether and can be
simplified by considering the satellites as point masses and the tether is a massless rod
whose bending, and stretching are negligible [4, 5]. In some other researches, the con-
necting rod is considered as a massive rigid body that provides more accurate dynamic
analysis [6, 7]. Although in the last years many interesting and theoretically reliable
results have been obtained, the lack of experimental validation is an important issue that
caught the eyes of the authors.
The main concern in this paper is to realize a detailed analysis of the behavior of
the articulated systems under space conditions. The obtained results can be served to
predict the behavior of the articulated satellite systems in different orbital conditions,
develop precise control schemes and supporting subsystems. Following these criteria,
it is possible to prevent failures or damages caused by oscillations/vibration during the
satellite operation. Although there exist some experimental banks which are used for
studying the tethered satellites, flexible space systems and space climber [8, 9] but until
now, they haven’t been any bench or results related to the experiment the articulated
bodies or dumbbell models. After the detailed mathematical modeling of the space
articulated systems, this work presents the necessary conditions and methods to carry
out the experimentation, validation, and control of articulated space systems in the ground
conditions.

2 Mathematical Model
2.1 Simplified Double Pendulum Model in Space Satellite

Consider the space structures as a pair of rigid pendulums with concentrated masses,
hence, a moment of inertia. This mechanism can be presented as a three-mass system
composed of the mother ship, which rotates around the Earth in an elliptical orbit, and
a pair of links (pendulums) connected to the main satellite by rotational joints as can be
seen in Fig. 1.
 Experimental Platform for Modeling and Control of Articulated Bodies 161

Fig. 1. Orbital model

Hereafter, the following terms and variables will be referred to as follows:

Jm : Joint with the mother ship.

J p : Joint between both pendulums,
θ1 (t) and θ2 (t): Libration angles of the center of mass of each body from the local
vertical,
L 1 : Distance from jm to the center of mass of the first pendulum,
L 2 : Distance from j p to the center of mass of the second pendulum,
L 1 : Distance between both joints,
L 2 : Full length of the second pendulum

The lengths L 1 , L 2 and L 1 are constant in this study. The orbital radius of the system
R = R(t)e2 is proposed as a variable for the cases of elliptical orders and related to the
orbital rate  = (t)e3 by using Kepler’s time equations. The orbital rotation reference
frame {e1 , e2 , e3 } has its origin in the center of the Earth. The resemblance of this model
with the experimental setup will be explained later.

2.2 Equations of Motion

The equations of motion presented in this study are obtained using Kane’s method [10].
The position of the center of mass of the first and second pendulum, are given by the
vectors R1 and R2 , as follows:

R1 = (L 1 sin θ1 )e1 + (R − L 1 cos θ1 )e2 (1)

   
R2 = L 1 sin θ1 + L 2 sin θ2 e1 + R − L 1 cos θ1 − L 2 cos θ2 e2 (2)

The generalized velocities are selected as

u 1 = θ̇1 (3)

u 2 = θ̇2 (4)
162 A. Hernandez-Maldonado et al.

Considering the effect of the orbital rate vector Ω, the velocity of the center of mass
of the first and second pendulum, V 1 and V 2 , are expressed respectively, as follows:
    
V 1 = (−R + L 1 cos θ1 )Ω̇ + L 1 θ̇1 cos θ1 e1 + Ṙ + L 1 sin θ Ω̇ + θ̇1 e2 (5)
   
V 2 = [ −R + L 1 cos θ1 + L 2 cos (θ1 + θ2 ) Ω̇ + L 1 cos θ1 + L 2 θ̇1 cos (θ1 + θ2 )
 
+ L 2 θ̇2 cos (θ1 + θ2 )]e1 + [ Ṙ + Ω̇ L 1 sin θ1 + L 2 sin (θ1 + θ2 )
+ L 1 θ̇1 sin θ1 + L 2 θ̇1 sin (θ1 + θ2 ) + L 2 sin (θ1 + θ2 )θ̇2 ]e2 (6)

The acceleration vectors of both the center of masses, A1 and A2 , are therefore

A1 = [−2 Ṙ Ω̇ + (R + L 1 cos θ1 )Ω̈ − L 1 (Ω̇ 2 sin θ1 + 2Ω̇ θ̇1 sin θ1 + sin θ1 θ̇12 − θ̈1 cos θ1 )]e1
+ [ R̈ + (−R + L 1 cos θ1 )Ω̇ 2 + 2L 1 Ω̇ θ̇1 cos θ1 + L 1 (Ω̈sin θ1 + θ̇12 cos θ1 + θ̈1 sin θ1 )]e2 (7)
 
A2 = [−2 Ṙ Ω̇ + −R + L 1 cos θ1 + L 2 cos (θ1 + θ2 ) Ω̈ − L 1 Ω̇ 2 sin θ1 − L 2 Ω̇ 2 sin (θ1 + θ2 )
− 2L 1 Ω̇ θ̇1 sin θ1 − 2L 2 Ω̇ θ̇1 sin (θ1 + θ2 ) − L 1 θ̇12 sin θ1 − L 2 θ̇12 sin (θ1 + θ1 )
− 2L 2 Ω̇ θ̇2 sin (θ1 + θ2 ) − 2L 2 θ̇1 θ̇2 sin (θ1 + θ2 ) − L 2 θ̇12 sin (θ1 + θ2 ) + L 1 θ̈1 cos θ1
+ L 2 θ̈1 cos (θ1 + θ2 ) + L 2 θ̈2 cos (θ1 + θ2 )]e1 + [ R̈ + (−R + L 1 cos θ1 + L 2 cos (θ1 + θ2 ))Ω̇ 2
+ L 1 Ω̈sinθ1 + L 2 Ω̈sin(θ1 + θ2 ) + L 1 θ̇12 cos θ1 + L 2 θ̇12 cos (θ1 + θ2 )
 
+ 2L 2 θ̇1 θ̇2 cos (θ1 + θ2 ) + L 2 θ̇22 cos (θ1 + θ2 ) + 2Ω̇ L 1 cos θ1 + L 2 cos (θ1 + θ2 ) θ̇1

+ L 2 θ̇2 cos (θ1 + θ2 ) + L 1 θ̈1 sin θ1 + L 2 θ̈1 sin (θ1 + θ2 ) + L 2 θ̈2 sin (θ1 + θ2 )]e2 (8)

The partial velocities of each body with respect to the generalized velocities are
∂ V1
= L 1 cos θ1 e1 − L 1 cos θ1 e2 (9)
∂u 1
∂ V1
=0 (10)
∂u 2
∂ V2  
= [L 1 cos θ1 + L 2 cos (θ1 + θ2 )]e1 + L 1 sin θ1 + L 2 sin (θ1 + θ2 ) e2 (11)
∂u 1
∂ V2
= [L 2 cos (θ1 + θ2 )]e1 + [L 2 sin (θ1 + θ2 )]e2 (12)
∂u 2
The gravitational forces affecting both pendulums are given by
R1
F g1 = −μm 1 (13)
|R1 |3
R2
F g2 = −μm 2 (14)
|R2 |3
Employing Kane’s equation,

K i + K i∗ = 0, i = {1, 2} (15)
 Experimental Platform for Modeling and Control of Articulated Bodies 163

The generalized active forces are considered as

∂V1 ∂V2
K i = F g1 · + F g2 · (16)
∂u i ∂u 1
And the generalized inertia forces, are expressed as
   
∗ ∂V1 ∂V2 ∂ω1 ∗ ∂ω2 ∗
Ki = · (m 1 A1 ) + · (m 2 A2 ) + T + T (17)
∂u i ∂u i ∂u i 1 ∂u i 2

Where ω1 , ω2 are the angular velocities of the first and second pendulum respectively,
while T1∗ , T2∗ are the inertial torque of each body defined as:

T i∗ = α i · I i − ω i × I i · ω i (18)

where α i , I i are the angular acceleration and the central inertia of the analyzed body.
Finally, the equations of motion can be obtained as

M q̈ + f = [0, 0]T . (19)

where M is the mass matrix, f is the nonlinear term vector combining the Coriolis and
centrifugal forces and q = [θ1 , θ2 ] is the state vector. The elements of M and f, are
expressed as follows:

M11 = −L 21 m 1 − L 22 m 2 − L 2 
1 m 2 − 2L 2 L 1 m 2 cos cos θ2 + I1 + I2
b3 c3
(20)

M12 = M21 = −L 22 m 2 − L 2 L 1 m 2 cos cosθ2 + I2c3 (21)

M21 = −L 22 m 2 + I2c3 (22)

  
f 1 = 2 L 1 m 1 + L 1 m 2 cos θ1 + L 2 m 2 cos (θ1 + θ2 ) L̇ Ω̇ − (L 21 m 2 + L 22 m 2 + L 2 1 m2
− R L 1 m 1 cos θ1 − R L 1 m 2 cos θ1 + 2L 2 L 1 m 2 cos θ2 − R L 2 m 2 cos (θ1 + θ2 ))Ω̈
 
− L 1 m 1 sin θ1 + L 1 m 2 sin θ1 + L 2 m 2 sin (θ1 + θ2 ) L̈ + (R L 1 m 1 sin θ1 + R L 1 m 2 sin θ1
μR(L 1 m 2 sin θ1 + L 2 m 2 sin (θ1 + θ2 ) μR(L 1 m 1 )
+ R L 2 m 2 sin (θ1 + θ2 ))Ω̇ 2 − 3
−
R3 R23
     
− m 2 (L 1 + L 2 cos θ2 ) L̇ 1 Ω̇ + θ̇1 + 2L 2 L 1 m 2 Ω̇sin θ2 θ̇2 + 2L 2 L 1 m 2 sin θ2 θ̇1 θ̇2
+ L 2 L 1 m 2 sin θ2 θ̇22 + I1b3 Ω̈ + I2c3 (23)
  
μL 2 m 2 L 1 sin θ2 − R sin (θ1 + θ2 )
f2 = + I2c3 Ω̈ − L 2 m 2 (−2 cos(θ1 + θ2 ) L̇ Ω̇ + L 2 Ω̈
R33
+ L 1 cos θ2 Ω̈ − Rcos (θ1 + θ2 )Ω̈ + L 1 Ω̇ 2 Sinθ2 ) + R̈ sin (θ1 + θ2 ) − R Ω̇ 2 sin (θ1 + θ2 )
 
+ 2L 1 Ω̇sin θ2 θ̇1 + L 1 sin θ2 θ̇12 + cos θ2 L̇ 1 Ω̇ + θ̇1 (24)
164 A. Hernandez-Maldonado et al.

2.3 Experimental Setup, Simplified Double Pendulum Model on Earth

The experimental setup, shown in Fig. 2, has been proposed to analyze the in-plane
motion of the bodies, by recreating the orbital conditions that affect the satellite behavior
in a ground-based environment.
The schematic representation and components used for this apparatus are shown in
Fig. 3. As it was previously mentioned, it isn’t possible to emulate, both the in-plane and
out-plane conditions on the ground, therefore, this study will focus on the first ones. The
proposed mechanism consists of an experimental platform with 3 degrees of freedom,
composed by an AC motor connected to a pulley that rotates (1) a turntable, which at the
same time supports an (2) inclined table. Finally, mounted parallel to the plate there’s a
pulley, connected to (3) a parallel slider which modifies the proximity of the analyzed
bodies to the center of the apparatus. The rotation of the main base resembles the orbital
motion, while the inclination and linear movements help to simulate more complex
trajectories like elliptical ones by modifying the orbital radius, as well as modifying
the gravitational effects on the bodies. In order to avoid twisting of electrical wires, all
the motors are energized through the slip ring located lower part of the experimental
setup and controlled via radio frequency by a pair of receiver-transmitter devices with a
commercial microcontroller and to monitor the position of the pendulums a camera will
be installed in the parallel slider (Fig. 4).

Fig. 2. Overview of the experimental apparatus.

One of the main differences between the ground-based experimental setup is gravity.
In orbital conditions, gravity depends on the orbital radius, while in the turntable it is
constant. These differences are reflected in the equations of motion for the experimental
setup and those of the orbital system, in terms of gravity and angular velocity. The effects
of gravity affecting the bodies in the inclined table are modified as follows

F g1 = −m 1 g(sin β e2 + cos β e3 ) (25)

F g2 = −m 2 g(sin β e2 + cos β e3 ) (26)

Where g is the gravitational acceleration constant on earth (g = 9.8 sm2 ) and β is the
inclination angle of the table. The angular velocity vector is now considered as

ω = ω sin β e2 + ω cos β e3 (27)

 Experimental Platform for Modeling and Control of Articulated Bodies 165

Fig. 3. Experimental platform

Fig. 4. Assembly of the articulated bodies.

Employing Kane’s equations as before, it’s possible to observe that the mass matrix
remains the same as in the orbital model, see Eq. (19), whereas the nonlinear terms in
the equations of motion for the experimental setup are now obtained as follows, where
I1b3 , I2c3 are the inertia of the bodies in the main axis:
    
f 1 = − g sin β L 1 m 1 + L 1 m 2 sin θ1 + L 2 m 2 sin(θ1 + θ2 ) + I1b3 ω̈ cos β − ω̇β̇ sin β

− L 1 m 1 2ω̇ Ṙ cos β cos θ1 + ω̈ cos β(L 1 − R cos θ1 ) + R ω̇β̇ sin β cos θ1
− L 1 ω̇β̇ sin β cos2 θ1 + R̈ sin θ1 − R ω̇2 cos2 β sin θ1 − L 1 ω̇2 cos θ1 sin θ1
  
+ L 1 ω̇2 cos2 β cos θ1 sin θ1 − L 1 ω̇β̇ sin β sin2 θ1 + I2c3 ω̈ cos β − ω̇β̇ sin β
 
− m 2 L 1 cos θ1 + L 2 cos(θ1 + θ2 ) − 2 Ṙ ω̇ cos β − R ω̈ cos β + L 1 ω̈ cos β cos θ1
+ L 2 ω̈ cos β cos(θ1 + θ2 ) + R ω̇β̇ sin β − L 1 β̇ ω̇ sin β cos θ1
− L 2 β̇ω sin β cos(θ1 + θ2 ) − L 1 ω̇2 cos2 β sin θ1 − L 1 ω̇ sin2 β sin θ1
166 A. Hernandez-Maldonado et al.

− L 2 ω̇2 cos2 β sin(θ1 + θ2 ) − L 2 ω̇ sin2 β sin(θ1 + θ2 )


− 2L 1 ω̇θ̇1 cos β sin θ1 − 2L 2 ω̇θ̇1 cos β sin(θ1 + θ2 ) − L 1 θ̇12 sin θ1
 
− L 2 θ̇12 sin(θ1 + θ2 ) + L̇ 1 cos θ1 ω̇ cos β + θ̇1 − 2ω̇θ̇2 L 2 cos β sin(θ1 + θ2 )

− 2L 2 θ̇1 θ̇2 sin(θ1 + θ2 ) − L 2 θ̇22 sin(θ1 + θ2 )
 
+ L 1 sin θ1 + L 2 sin(θ1 + θ2 ) ( R̈ − cos2 β(R − L 1 cos θ1
− L 2 ω̇2 cos(θ1 + θ2 ) + L 1 ω̈ cos β sin θ1 + L 2 ω̈ cos β sin(θ1 + θ2 ) + L̇ 1 θ̇1 sin θ1

+ L 1 θ̇12 cos θ1 + L 2 θ̇12 cos(θ1 + θ2 ) + 2L 2 θ̇1 θ̇2 cos(θ1 + θ2 ) + L 2 θ̇22 cos(θ1 + θ2 )
  
+ ω̇ L̇ 1 cos β sin θ1 − β̇ sin β L 1 sin θ1 + L 2 sin(θ1 + θ2 )

+ 2 cos β((L 1 cos θ1 + L 2 cos(θ1 + θ2 ) θ̇1 + L 2 θ̇2 cos(θ1 + θ2 ) (28)
 
f 2 = − gL 2 m 2 sin β sin(θ1 + θ2 ) + I2c3 cos β ω̈ − ω̇β̇ sin β
1 
− L 2 m 2 −16 Ṙ ω̇ cos β cos(θ1 + θ2 ) + 8L 2 ω̈ cos β + 4L 1 ω̈ cos(β − θ2 )
8

− 4R ω̈ cos(β − θ1 − θ2 ) + 4L 1 ω̈ cos(β + θ2 )
− 4R ω̈ cos(β + θ1 + θ2 ) − 8L 2 ω̇β̇ sin β − 4L 1 ω̇β̇ sin(β − θ2 ) − L 1 ω̇2 sin(2β − θ2 )
− L 1 ω̇2 sin(2β − 2θ1 − θ2 ) + 4R ω̇β̇ sin(β − θ1 − θ2 ) − L 2 ω̇2 sin(2(β − θ1 − θ2 ))
+ 2R ω̇2 sin(2β − θ1 − θ2 ) + 6L 1 ω̇2 sin θ2 − 4L 1 ω̇β̇ sin(β + θ2 ) + L 1 ω̇2 sin(2β + θ2 )
+ 8 R̈ sin(θ1 + θ2 ) − 4R ω̇2 sin(θ1 + θ2 ) − 2L 2 ω2 sin(2(θ1 + θ2 )) + 4R ω̇β̇ sin(β + θ1 + θ2 )
+ L 2 ω̇2 sin(2(β + θ1 + θ2 )) − 2R ω̇2 sin(2β + θ1 + θ2 ) − 2L 1 ω̇2 sin(2θ1 + θ2 )
 
+ L 1 ω̇2 sin(2β + 2θ1 + θ2 ) − 8L 1 ω̇θ̇1 sin(β − θ2 ) + 8L 1 ω̇θ̇1 sin(β + θ2 ) + 8L 1 θ̇12 sin θ2
 
+ 8 L̇ 1 cos θ2 ω̇ cos β + θ̇1 (29)

One form to obtain the inclination angle of the turn table that will emulate the
gravity effects of the orbit system in the experimental setup is by relating the gravidity
gradient torque on the orbit with the one obtained in the ground-based platform [7], so
the inclination angle will be obtained as follows:

−g + g 2 + 4ω4 (R − L)(R + 3L)

β = sin−1 (30)
2(R − L)ω2

In order to simplify the analysis for elliptical orbits, this angle can also be
obtained through algebraic relations between the gravity force in orbit (Fge ) and in
the experimental setup (Fg p ,) as follows:

Gm e
Fge = m s , Fg p = g sin β m s (31)
r2
Fge = Fg p (32)

Gm e
β = sin−1 (33)
gr 2
 Experimental Platform for Modeling and Control of Articulated Bodies 167

Where G is the universal gravitational constant, m e the Earth’s mass, r the orbital
radius to be emulated and g = 9.8 sm2 .

3 Results of Numerical Simulations

In order to validate the equations of motion obtained for the orbital system (Eqs. 20–24)
and the experimental setup (Eq. 28–29), some numerical simulations where made. It
is assumed that the orbital radius R of the mother ship and the semimajor axis for the
case of elliptical orbits are 6800, with an eccentricity of e = 0.025 for the last case. The
orbital period is then 5580 s (93 min).
As for the experimental setup, the parameters are set as follows. The weight of each
pendulum is 20.55 g, the lengths, L 1 and L 2 , are both 100 mm while L 1 is 120 mm.
The distance R between the end of the reel system and the main shaft along the inclined
table is 70 cm. In the case of circular orbits (e = 0) the angular rotation rate of the main
shaft is set as 0.012, so the period evaluated is 50 s and the inclination is 0.011°. As for
the case of the elliptic orbit, the inclination angle is evaluated as a function that depends
on the orbital radius which is variable over time (Figs. 5 and 6).

Orbital model Experimental platform

A

Fig. 5. Numerical results for circular orbit: (a) First pendulum angle θ1 , and (b) second pendulum
angle θ2 .
168 A. Hernandez-Maldonado et al.

Orbital model Experimental platform

C

Fig. 6. Numerical results for circular orbit: (a) First pendulum angle θ1 , and (b) second pendulum
angle θ2 .

4 Conclusion
From the graphics it is possible to observe that the mathematical model for the behavior
of orbital system simplified as a group of pendulums and from the proposed experimen-
tal setup are similar, but since the system is very sensitive to any perturbation there are
still some slight differences produced by the parameters considered in the experimen-
tal platform, such as the position of the center of mass in each pendulum and inertia.
Therefore, some adjustments and considerations need to be done in order to increase the
accuracy and precision of the experimental setup, as well as the real-life experiments
and programming of all the components. Nevertheless, the simulations prove the math-
ematical model considered in this paper gives an optimal approximation to analyze the
behavior and control of articulated bodies in orbital conditions.

References
1. Menon C, Bombardelli C (2007) Self-stabilising attitude control for spinning tethered
formations. Acta Astronaut 60(10–11):828–833
2. NASA Orbital Debris Program Office (2013) Monthly number of objects in earth orbit by
object type. Orbit Debris Q News 17(1):8
3. European Space Agency January 2019 «Space debris by the numbers» European Space
Agency. https://www.esa.int/Safety_Security/Space_Debris/Space_debris_by_the_numbers.
Accessed June 2019
4. Tyc G (1999) Dynamics and stability of spinning flexible space tether systems
5. Misra A, Xu D, Modi V (1986) On vibrations of orbiting tethers. Acta Astronaut 13(10):587–
597
6. Fritzkowski P, Kaminski H (2008) Dynamics of a rope as a rigid multibody system. J Mech
Mater Struct 3(6):1059–1075
 Experimental Platform for Modeling and Control of Articulated Bodies 169

7. Kaminski H, Fritzkowski P (2013) Application of the rigid finite element method to modeling
ropes. Latin Am J Solids Struct 10(1):91–99
8. Kojima H, Iwasaki M, Fujii HA, Blanksby C, Trivalio P (2004) Nonlinear control of libration
motion of tethered satellites in elliptics orbits. J Guid Control Dyn 27(2): 229–239
9. Kojima H, Sugimoto Y, Furukawa Y (2011) Experimental study on dynamics and control of
tethered satellite systems with climber. Acta Astronaut (69): 98–108
10. Kane TR (2005) Dynamics: theory and applications. McGraw-Hill Series in Mechanical
Engineering, New York
 Design and Control of a Passive Solar
Tracking System Using a Sky Imager

M. Angulo1 , A. Dı́az-Ponce1(B) , L. Valentı́n1 , R. Valdivia2 ,

and S. Keshtkar2
1
CONACYT-Centro de Investigaciones en Óptica, A.C., Aguascalientes, Mexico
{mangulo,adiaz,luismvc}@cio.mx
2
Instituto Tecnológico y de Estudios Superiores de Monterrey, Tamaulipas, Mexico
{r.valdivia,skeshtkar}@tec.mx
https://www.cio.mx/investigacion/energia solar/

Abstract. Solar energy is considered as one of the most important

sources of renewable energy due to the possibility of being converted
into electrical power, which could result in a feasible approach to global
energy problems. Concentrated Solar Power (CSP) and High Concen-
tration Photovoltaics (HCPV) systems present an alternative by the use
of cost-efficient concentrating optics. However, these systems require a
high-precision solar tracking system, allowing to stay perpendicular to
the sun rays in order to achieve its correct functioning. In this work, a
methodology for the detection and monitoring of the apparent movement
of the sun through a passive image acquisition system is presented. The
setup operates by processing sky dome images to find the solar-center.
The development cost of this device is relatively low, it does not require
articulated elements, can detect the presence of clouds, and could be
compatible with multiple CSP systems. Preliminary results of a two-axis
solar tracking by locating position of the sun in the sky with the proposed
system are depicted.

Keywords: Concentrating solar power · Image processing ·

Sky-imaging system · Solar photovoltaic · Sun detection · Solar
tracking system

1 Introduction

In the last two decades, renewable alternatives have been considered as an impor-
tant source of energy. The axiomatic depletion of fossil fuels and global warming
have encouraged the development and implementation of systems that generate
energy from renewable sources.
Solar energy has protruded by a popular approach and is implemented in
many nations of the world compared to conventional power sources. In addition,
it is considered as an unlimited reserves, it has no cost, it is pollution-free, among
other multiple benefits [1–3].

c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 170–178, 2020.
https://doi.org/10.1007/978-3-030-45402-9_17
 Design and Control of a Passive Solar Tracking System Using a Sky Imager 171

Solar technologies transform solar radiation into another type of energy,

which can be exploited by the user. Typically, this kind of technology has been
implemented by the photovoltaic (PV) industry and the CSP industry in regions
where there are scarce clouds, haze and, of course, where sun hours are abun-
dant [4,5]. It is important to mention that these configurations achieve their
optimum efficiency when the sun rays impact perpendicular to the surface of
the converting part of the setup, such as thermal solar collectors, solar cells,
among others [6–9]. This is possible due to a solar tracking system. Eventually,
the power generation and performance of solar systems decrease dramatically if
there is a significant solar orientation error, therefore, it is important to keep
the error as low as possible.
There are several configurations of solar tracking systems in the literature,
classified based on their degrees of freedom (DoFs) and/or their performance.
The former can follow the apparent sun position just in one-axis and, for greater
accuracy, there are two-axis solar tracking systems [10–12]. The two-axis solar
tracking can be either: polar tracking (equatorial) or azimuth/elevation tracking
(altitude-azimuth).
In terms of the performance, the tracking systems can be classified as pas-
sive or active. In the same way, active trackers can be classified into two types:
open and closed loop [13]. A passive tracker system works without electronic
sensors, actuators or electronic control systems. On the other hand open loop
systems, works through mathematical model to predict the apparent sun posi-
tion and use the time and geographical position in which the system is installed.
Unfortunately, the angular position of the actuator that is being controlled can
not be known, so it is impossible to re-adjust any positioning error caused
by a disturbance [14–16]. The closed-loop systems employ, typically, the infor-
mation of electro-optical sensors to orient the tracking mechanism and esti-
mate the location of the sun. The most used are photoresistors or photodi-
odes arrangements, complementary metaloxide-semiconductor (CMOS), micro-
electro-mechanical system (MEMS), pixel sensor (image sensor), among others.
These sensors have different advantages and disadvantages and their selection
depends on solar technology, cost-effectiveness, the requirements of the system
of interest and the strategies of commercialization [17]. The main disadvantages
of this sensor are: its spectral response changes with respect to temperature or
radiation variations, its response can presents areas of non-linearity, it has low
repeatability, it can be easily damaged due to the humidity and dust and suffer
degradation over time.
One of the most important parameters in solar sensors is the field of view
(FOV) – a parameter which allows predicting the possible angle of sun position.
Typically, the wider the field of view of a sensor, the lower the accuracy of the
solar location (FOVb in Fig. 1). This phenomenon can be explained by the fact
that the sensor is susceptible to the measurement of noise caused by diffuse
radiation in the environment. On the other hand, a narrow field of view (FOVa
in Fig. 1) allows to a achieve higher accuracy by capturing only direct solar
radiation; however, when the sun is out of range, it can not be located without
a pre-location, which requires a more complex control system.
172 M. Angulo et al.

FOVa
FOVb

Sensor

Fig. 1. FOV demonstration of two solar sensors.

In this work, a low-cost Passive Sun Tracking System (PSTS) based on image
processing is proposed. The developed methodology depend on the implementa-
tion of several algorithms, which allow to identify and evaluate the apparent sun
location with a FOV of 180◦ . The proposed system was tested in Aguascalientes
city, Mexico (21◦ 51'N 102◦ 18'W).

2 Methodology

The experimental schematic representation of the system is shown in Fig. 2,

this system is composed of two parts: the PSTS and the solar tracking system
detailed in Sect. 2.1 and 2.3 respectively. The operation of the system begins
with the capture of the solar rays reflected from the convex mirror towards the
Raspberry-Pi camera of the PSTS.
In order to compute the apparent position of the sun, the image that was
acquired is processed. Next, the Zenith and Azimuth angles are sent to the solar
tracker controller. This procedure is done with a frequency set by the user. The
control system of the solar tracking system takes as reference the received Zenith
and Azimuth angles and, finally, the controller drives the actuators until they
reach the desired position.

2.1 Passive Sun Tracking System

The PSTS developed consists of a low-cost sky imager device, image acquisition,
processing algorithms and the support structure (See Table 1). The prototype
is based on a Raspberry Pi Camera module V2 that uses a CMOS IMX219PQ
sensor with 8-megapixel native resolution. The camera is protected by a case
particularly designed for outdoor applications able to protect it from rain and
dust. The convex mirror is mounted in a structure built by 20 mm structural
aluminum using an upward facing configuration that achieves a FOV of 180◦
(See Fig. 3) , which allows to visualize the apparent sun position at any time of
the day. Furthermore, in order to stabilize the camera, a set of clamping cables
have been fastened from a point near the camera housing to the structure base.
Design and Control of a Passive Solar Tracking System Using a Sky Imager 173

PSTS Image processing Two axis solar tracking

Camera
Image
RGB to HSL Image binarization

Evaluating the
Sun distance between Minimum enclosing
rays Convex objects found and circle estimation
mirror dome center

Solar tracking system

Apparent sun control algorithm
position Azimuthal and
zenithal angle

Fig. 2. Schematic of experimental setup.

Table 1. Cost of the materials used in the construction of the prototype.

Description Quantity Cost [$ USD]

Structural Al. 20 mm 6m 38.5
Convex mirror 18 1pc 70.0
Camera board V2 1pc 23.0
Raspberry Pi 3 B 1pc 77.0
Miscellaneous materials 33.0
Total 245.5

Raspberri-Pi camera

Support structure
Hook tensor

Convex mirror
Structure base

Fig. 3. PSTS designed.

174 M. Angulo et al.

Although the system needs to know neither its geographical loca-

tion(longitude, latitude or altitude) nor the date and time, it requires a pre-
orientation with respect to the solar tracker. The system can be installed on
irregular surfaces, but it is necessary to perform a configuration routine in order
to position the plane of the normal dome towards the zenith. In addition, the
system has a low development cost due to some advantages such as the inclusion
of economic material and the absence of moving parts. It is important to mention
that the setup does not have sensors which are likely to suffer deterioration over
time and change their linearity or their spectral response. On the other hand,
a single prototype is able to command the position of several single or two-axis
tracking systems.
One of the main advantages of using this kind of sensor in comparison of pho-
todiodes or light-dependent resistors is its low sensitivity to weather conditions
such as temperature, humidity and diffuse radiation from the environment. To
achieve optimal solar tracking, It is possible to adjust the detection frequency of
the sun position, and know the state of cloudiness in the day. The information
is sent wirelessly to a user interface, where the apparent position of the sun,
obtained with the developed prototype, can be displayed in real-time.

2.2 Sun Position Detection Process

In order to detect and monitor the apparent position of the sun, an image-
based sun detection algorithm has been developed. The sun detection process
begins with the acquisition of an image in an RGB color space, then a color
space conversion from RGB to HSL color space is performed. HSL is another
color space which represents the color on three channels: hue, saturation, and
brightness (how bright the color is). Thus, objects with very large brightness
values can be easily separated from the background in the image; it is assumed
that the center of the brightest region in the sky image represents the position
of the solar center. In this sense, to distinguish the sun from the rest of the
elements in the image, a binarization of the image is performed. Figure 4 shows
the resulting image after the binarization process, where the segmented objects
are those with the highest brightness.

Fig. 4. Binarization of the image. Figure 4a shows the input color image. Figure 4b
shows the result of the binarization.
 Design and Control of a Passive Solar Tracking System Using a Sky Imager 175

From each object found, the minimum enclosing circle to get the center, as
well as the radius, is computed. To discriminate the sun from all those objects
found in this process, a distance to the dome center (which has also been
detected) is calculated. If the distance of an element is lower than the dome
radius then, the algorithm considers this element as the circle that represents
the sun. A sun detection example of this process is shown in Fig. 5.

Fig. 5. Sun detection example.

2.3 Solar Tracker

To verify the proposed passive solar location strategy, a two-axis solar tracker
is used. This mechanism operates with a slewing drive that provides torque,
accuracy, precision, efficiency, and reliability in solar tracking (see Fig. 6).
Once the developed system locates the apparent position of the sun with
respect to its reference frame, the azimuthal and zenithal angles are determined,
which will then be sent to an Arduino controller. These angular values are inter-
preted by the solar tracking system control algorithm as input signals. In subse-
quent work, the characterization of solar tracking accuracy will be carried out.

3 Results and Discussion

A graphical user interface (GUI) was the developed in which the user can observe
relevant information, such as the zenith and azimuth angles sent to the tracker
system, the current image of the sky with the sun detected, among other infor-
mation. Additionally, the user has the possibility of using the interface to make
modifications of some of the parameters directly related to the detection of the
sun position. Figure 7 shows the start window of the GUI; from this window the
user is able to configure the initial system parameters.
In order to carry out the final solar tracking tests, it is necessary to instrument
the PSTS with a magnetometer to geographically orientating the system. Once
its geographical orientation is determinated, it will be sent wirelessly to the
solar tracker controller, where a coordinate transformation is required to drive
the actuators and align the system to the apparent position of the sun.
176 M. Angulo et al.

Fig. 6. Two-axis solar tracking system.

Fig. 7. Interface start window.

On the other hand, Fig. 8 shows the system GUI, where it is possible to
observe the detection of the apparent position of the sun (performed by the
proposed system), in addition to other relevant information such as geographical
location, time, date, etc.
As mentioned in this document, the proposed PSTS has several advantages
over the tracking sensors commonly used in the literature; however, it is necessary
to measure the final tracking error to determine its viability in CSP and HCPV
systems.
 Design and Control of a Passive Solar Tracking System Using a Sky Imager 177

Fig. 8. System graphical user interface.

4 Conclusions
In this work, solar tracking algorithm was developed through the implementation
of a novel passive solar location strategy. The estimated cost of the prototype,
without considering the solar tracking system, was $ 245.5 USD, which makes it
economically viable to be implemented in low-cost solar applications that require
solar tracking. One of the biggest advantages of the system developed is that it
reaches a FOV of 180◦ ; that is, it is possible to locate the position of the sun at
any time of the day, requires low maintenance, has no moving parts and does not
use sensors that deteriorate over time so the efficiency of the system does not
decrease. In addition, it is important to mention that the PSTS is not affected
by diffuse radiation; therefore, the accuracy of solar detection of the system does
not decrease on sunny and cloudy days. However, in days with severe cloudiness
the sun position can not be localized.
As future work, the characterization of the solar tracking system will be
performed in order to know the tracking precision; nevertheless, it is essential to
mention that there is no official procedure for the calibration of the tracking error
measurement tool established by the International Electrotechnical Commission
(IEC) or some other entity.

References
1. Sobri S, Koohi-Kamali S, Rahim N (2018) Solar photovoltaic generation forecasting
methods: a review. Energy Convers Manage 156:459–497 ISSN 0196-8904
2. Herche W (2017) Solar energy strategies in the US utility market. Renew Sustain
Energy Rev 77:590–595 ISSN 1364-0321
3. Alemán-Nava G, Casiano-Flores V, Cárdenas-Chávez D, Dı́az-Chavez R, Scarlat
N, Mahlknecht J, Dallemand J, Parra R (2014) Renewable energy research progress
in Mexico: a review. Renew Sustain Energy Rev 32:140–153 ISSN 1364-0321
178 M. Angulo et al.

4. Kabir E, Kumar P, Kumar S, Adelodun A, Kim K (2018) Solar energy: potential

and future prospects. Renew Sustain Energy Rev 82, Part 1:894–900 ISSN 1364-
0321
5. Renewables 2013 (2013) Global Status Report 2013, Network R
6. Carballo J, Bonilla J, Berenguel M, Fernández-Reche J, Garcı́a G (2019) New
approach for solar tracking systems based on computer vision, low cost hardware
and deep learning. Renew Energy 133:1158–1166 ISSN 0960-1481
7. Azizi K, Ghaffari A (2013) Design and manufacturing of a high-precision sun track-
ing system based on image processing. Int J Photoenergy 2013:7 754549
8. Kasaeian A, Nouri G, Ranjbaran P, Wen D (2018) Solar collectors and photo-
voltaics as combined heat and power systems: a critical review. Energy Convers
Manage 156:688–705 ISSN 0196-8904
9. Enrigh, JP, Godard (2008) Design optimization of a digital sun sensor for use with
parametric processing. IEEE Trans. Instrum. Meas. 57(10):2188–2195
10. Mousazadeh H, Keyhani A, Javadi A, Mobli H, Abrinia K, Sharifi A (2009) A
review of principle and sun-tracking methods for maximizing solar systems output.
Renew Sustain Energy Rev 13(8):1800–1818
11. Singh R, Kumar S, Gehlot A, Pachauri R (2018) An imperative role of sun trackers
in photovoltaic technology: a review. Renew Sustain Energy Rev 82, Part 3:3263–
3278 ISSN 1364-0321
12. AL-Rousan N, Mat Isa N, Mat Desa M (2018) Advances in solar photovoltaic
tracking systems: a review. Renew Sustain Energy Rev 82, Part 3:2548–2569 ISSN
1364-0321
13. Huynh DC, Nguyen TM, Dunnigan MW, Mueller MA Comparison between open-
and closed-loop trackers of a solar photovoltaic system. In: 2013 IEEE Conference
on Clean Energy and Technology (CEAT), pp 128–133 IEEE
14. Nsengiyumva W, Chen S, Hu L, Chen X (2018) Recent advancements and chal-
lenges in Solar Tracking Systems (STS): a review. Renew Sustain Energy Rev 81,
Part 1:250–279 ISSN 1364-0321
15. Chia-Yen L, Po-Cheng C, Che-Ming C, Chiu-Feng L Sun tracking systems: a
review. Sensors (Basel) 9:3875–3890 https://doi.org/10.3390/s90503875
16. Lynch W, Salameh Z (1990) Simple electro-optically controlled dual-axis sun
tracker. Sol Energy 45(2):65–9
17. Salgado-Conrado L (2018) A review on sun position sensors used in solar applica-
tions. Renew Sustain Energy Rev 82, Part 3:2128–2146 ISSN 1364-0321
 Attenuation of the Amplitude Vibration
Level by Using Nonlinear Active Control
Techniques Applied to a Hybrid Short
Journal Bearing Using a Particular
Rotor-Bearing System Configuration

D.-E. Rivera-Arreola1 , J.-C. Gómez-Mancilla2 , T. Hernández-Cortés1(B) ,

and H. F. Abundis-Fong3
1
Univ. Politéc. de Pach., Carretera Ciudad Sahagún-Pachuca Km. 20,
43830 Pachuca, Hidalgo, México
{eduardorivera,tonatiuh}@upp.edu.mx
2
Inst. Politéc. Nal., Av. Luis Enrique Erro S/N,
Unidad Profesional Adolfo López Mateos, Zacatenco Alcaldı́a Gustavo A. Madero,
C.P. 07738 Ciudad de Mexico, Mexico
gomezmancilla@gmail.com
3
TecNM/I.T. Pachuca, Felipe Angeles Km. 84.5, Venta Prieta,
42083 Pachuca de Soto, Hidalgo, Mexico
habundis@itpachuca.edu.mx

Abstract. Inherent to turbomachinery are some common operational

problems, such as the critical amplitude of motion that the machine may
experience when passing through one or more of its natural frequencies.
High range of vibration levels can be catastrophic to the equipment and
its attenuation is needed to keep the system working healthily. To that
end, the present work proposes the use of nonlinear active control tech-
niques applied to a single hybrid short journal bearing using a particular
rotor-bearing system configuration, which not require the assumption of
an exact symmetry in the operation of the two journal bearings located
at both ends of the rigid shaft such as the classical Jeffcott rotor model.
The analysis is focused by reducing the amplitude shaft vibration of the
rigid rotor crossing its first resonance. The oil injection port is modeled
by mean of the Dirac Delta Function.

Keywords: Hybrid-short journal bearing system · PID controller ·

Reynolds equation · State-space · Dirac delta function

1 Introduction
Turbomachinery is currently used across several industries because its technology
can be applied in a wide range of rotating equipment, such as aircraft engines,
wind turbines, and other forms of transfer energy between a rotor and a fluid.
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 179–192, 2020.
https://doi.org/10.1007/978-3-030-45402-9_18
180 D.-E. Rivera-Arreola et al.

Thus, the health of rotating machinery is crucial. Some resonance might occur
within the machine’s range of operations. Therefore, it is of paramount impor-
tance to identify the critical speeds of the rotor, to prevent the turbo rotor
from operating at such speeds. Hydrodynamic bearings are commonly used to
attenuate high vibration levels due to because they behave with incomparable
smoothness. However, the problem of self-excited vibration, called fluid-induced
vibration, on the thin film in the rotating components may lead to large ampli-
tudes of motion and possible damage to the rotor [1,2]. Externally pressurized
bearings have been developed and successfully used in machinery that operates
under high-speed and heavy-load conditions. Hybrid bearings combine the phys-
ical mechanisms of both hydrodynamic and hydrostatic bearings. Some advan-
tages of hybrid bearings over purely hydrodynamic bearings are that wear and
tear are avoided when the shaft starts and stops, eliminating metal to metal
contact and allowing for very low starting torque. Attractive features include
tolerance of wider variations in bearing clearance, good bearing film stiffness
at zero rotating speed and the ability to employ higher viscosity oils at high
speed. In this sense, hybrid bearings are designed to take maximum advantage
of hydrodynamic and hydrostatic effects and, when designed appropriately, have
superior performance at high speeds [3,4]. Donald-Bently Corp. USA [5] carried
out experimental studies related to an externally pressurized fluid-film bearing
mainly focused on the diagnosis of machinery malfunction. The most important
conclusions can be summarized as follows:
– When externally pressurized, the rigidity of the bearings increases remarkably.
– Pressurization can produce dynamic instability.
– Under certain conditions the external pressurization might modify the reso-
nance speeds.
– Pressure variation can allow adequate control of the rigidities in the bearing
and can be implemented in more than one injection port.
The theory of fluid film lubrication is based mainly on the work of Osborne
Reynolds (1842–1912). Under ideal conditions, the Reynold’s equation is widely
used to compute the film pressure for hydrodynamic journal bearings [6–12].
Over the last few decades, several numerical approaches have been used, such
as the multigrid techniques [13], the finite element methods [14,15] or simpler,
faster yet still accurate methods like the mobility method [16] or the impedance
method [17,18]. In general, different strategies are employed to perform analy-
sis of hydrodynamic journal bearings. Nowadays, with the fast development in
technology, combining mechanics and mechatronics has given a great impulse
to the development of the turbomachinery. This interdisciplinary knowledge has
allowed the use of new electromagnetic devices as well as the application of con-
trol laws which can be useful and practical for the attenuation in highs levels
of vibration. From this point of view Nicoletti et al. applied some linear and
nonlinear control techniques to tilting-pad journal bearings. Theoretical results
showed that the proper selection of the controllers can be used to control the
oil supply to reduce the level of vibration amplitude of the rotor-bearing system
[19]. Santos et al. derived theoretical expressions considering the L/D = 1 ratio
 Attenuation of the Amplitude Vibration Level 181

for multirecess hybrid journal bearings with active lubrication. Analytical results
of the bearing dynamic coefficients calculations were presented [20]. Alejandro
et al. studied active lubrication applied to tilting-pad journal bearings. The
equations governing the actively lubricated bearings was derived by including
the pipelines dynamics effects into the mathematical models [21]. Arias-Montiel
et al. applied a Linear Quadratic Regulator (LQR) to reduce the vibration ampli-
tudes in a rotor system with active control scheme. Their results demonstrated
significant reductions in the unbalance response [22]. Jorge et al. studied active
tilting-pad journal bearings supported by flexible rotors. A reduction about 30%
were achieved around the resonant zone with hybrid lubrication [23]. Leonardo
Biagiotti et al. studied the reduction of a rotor vibration amplitude using PID
tuning methods. Five tuning methods were used and compared for the PID. The
vibration amplitude of the critical speed was reduced and the system damping
was increased for every tuning method [24].
In the present work, the proposed configuration provides an analysis by using
a single hybrid short journal bearing with four oil injection ports. Thus, does
not require the assumption of an exact symmetry in the operation of the two
journal bearings located at both ends of the rigid shaft. The hybrid bearing
is modeled ad hoc by the termed closed solution to the Reynolds Lubrication
Equation, and the Dirac Delta Function is introduced in order to model the oil
injection port. Also, this work will incorporate a PID controller using the Ziegler-
Nichols (ZN) tuning method applied to the nonlinear analytical expression to
eliminate and can cope with unexpected operating conditions. To determine
the stability of the nonlinear equation of motion and thus ensure the optimal
functionality of the PID controller, the mathematical tools related to the theory
of Lyapunov’s stability criterion are used. Finally, the governing equations of
motion corresponding to the proposed physical system are given to validate our
experiments.
The remainder of the paper is organized as follows: a formulation and descrip-
tion of the mathematical models are presented in Sect. 2, the study an implemen-
tation of control laws are analyzed and studied in Sect. 3, the tests are presented
in Sect. 4 and, the conclusion are given in Sect. 5.

2 Mathematical Modeling of the Rotor-Bearing System

The rotor-bearing configuration studied in the present work is a system based on
a rigid shaft and two equidistant disks supported by only one hybrid short journal
bearing located in the middle of the two disks. Figure 1 depicts a schematic
diagram of the physical rotor-system configuration and the stiffness and damping
coefficients. Where ω is the shaft speed considering a counterclockwise rotation,
the force Ax is generated by the imbalance and weight of the disk 1, on X-
axis and it is located through. Besides, Ay is the disk imbalance force in the
Y -axis; forces on Bx and By are generated by oil pressure in the hybrid short
journal bearing located through the segment. Similarly, force Cx is caused by the
imbalance and weight of the disk 2, on the X-axis and Cy is the disk imbalance
182 D.-E. Rivera-Arreola et al.

force in the second disk on the Y -axis, located at OA = l. To simplify the

mathematical expressions, some considerations must be assumed:

Fig. 1. Rigid shaft model and schematic repre- Fig. 2. Cross section for a journal
sentation of the cross section considering stiff- bearing with nomenclature.
ness and damping coefficients.

1. The shaft is Rigid.

2. The disks are equidistant to the bearing.
3. The moment of inertia of the shaft is ignored since it is much smaller than
the moment of inertia of the disks.

The equations of motion presented below are derived by Newton-Euler formula-

tion, where the angular displacements generated by the shaft are described by
θx and θy respectively.
 
I0 θ̈x = Mxz , I0 θ̈y = Myz (1)

Rewriting Eq. (1) and taking into account the equilibrium system based on [25],
results.
  
9
I0 θ̈x + l2 CXX θ̇x (1 + tan2 θx ) + CXY θ̇y (1 + tan2 θy )
16
+ KXX tan θx + KXY tan θy ] =
l
md1 (ω 2 ex1 cos(ωt) − αey1 sin(wt)) + lmd2 (ω 2 ex2 cos(ωt) − αey2 sin(wt)) (2)
2
  
9
I0 θ̈y + l2 CY X θ̇x (1 + tan2 θx ) + CY Y θ̇y (1 + tan2 θy )
16
+ KY X tan θx + KY Y tan θy ] =
l
md1 (ω 2 ex1 sin(ωt) + αey1 cos(wt)) + lmd2 (ω 2 ex2 cos(ωt) + αey2 cos(wt)) (3)
2
where:   2  
1 2 l 1 2 2
IO = md1 rd1 + md1 + md2 rd2 + md2 (l) (4)
4 2 4
 Attenuation of the Amplitude Vibration Level 183

is the mass moment of inertia, [kg ·m2 ]; α is the angular acceleration of the shaft,
[ rad x x
s2 ]; l is the length of the shaft, [m]; md1 is the disk’s mass [kg]; e1 = e2 =
y y
e1 = e2 are the imbalance distance of the disk masses in the x and y direction,
W
[m]; kij = C r
Kij and cij = CWrω
Cij are the dimension stiffness and damping
coefficient [ m ] and [ m ] respectively in the direction i, j where, W = md · g is
N Ns

the weight of the system, [N ], Cr is the radial clearance, [m]. Finally, Φ is the
equilibrium attitude angle [rad], see Fig. 1 and 2. To compute the values of Kij
and Cij the interested reader can consult references [26,27].

2.1 Reynolds Equation and Boundary Conditions

In lubrication theory, the Reynold’s equation is a partial differential equation

expressed in terms of pressure. Once the pressure profile is evaluated, all other
bearing performance parameters can be determined. According to the movement
of the shaft, the pressure field within a journal bearing is generated by the
formation of a lubricant wedge, depicted in Fig. 2. Hence, it is of the utmost
importance to determine the location of the maximum pressure and the regions
where the lubricant oil film can produce cavitation [28,29]. The Reynolds theory
of lubrication can be obtained based on the assumptions of [10,11]:

1. The fluid is assumed to be Newtonian.

2. Inertia and body forces terms are assumed to be negligible compared to the
viscous terms.
3. Variation of pressure across the film is assumed to be negligible, i.e., ∂P
∂z = 0,
so P = (x, y) only.
4. Flow is laminar.
5. Curvature effects are negligible. This implies that the thickness of the lubri-
cant film is much smaller than the length or width of the bearing so that the
physical domain of the flow can be unwrapped, allowing for the use of the
cartesian coordinate system.

   
∂ ∂P ∂ ∂P
h3 + R2 h3 =
∂θ ∂θ ∂z ∂z
µR2  ω
12π 3 Cr ε̇ cos θ + Cr ε φ̇ − sin θ (5)
Cr 2

L L
− ≤z≤ , 0 ≤ θ ≤ 2π, h(π) = 1 + ε cos θ
2   2  
L −L
P = 0, P , P (θ + 2π) = P (θ)
2 2

where, P is the pressure, [P a]; θ is the circumferential coordinate, [rad]; z is

the axial coordinate, [m]; φ is the equilibrium angle (attitude), [rad]; Cr is the
184 D.-E. Rivera-Arreola et al.

radial clearance, [m]; e is the eccentricity, [m]; ε is the dimensionless eccentric-

ity, ε = Cer ; R is the journal radius, [m]; RB is the bearing radius, [m]; µ is the
dynamic viscosity, [P a · s]; H is the thickness of the oil film, [m]; h is dimen-
sionless thickness of the oil film, h = CHr ; w is the angular speed considering a
counterclockwise rotation, [rad/s]; D is the journal bearing diameter, [m]; and
L the journal bearing length, [m]. Also, this work is restricted to short length
bearings, i.e. L << D. As a consequence, the pressure gradients, ∂P ∂P
∂θ << ∂z ,
allow us to ignore the first term in Eq. (5). Further, a short bearing in steady-
state (ε̇ = φ̇ = 0), and by using the following dimensionless variables: z = L2 ,
2  R 2
p
p = Pdim and Pdim = µN CRr = µ 2π ω
Cr the Reynolds equation for a
hydrodynamic journal bearing expressed in dimensionless form provides:
   2
∂ 3 ∂p L
h = 12π (ε sin θ) (6)
∂z ∂z D

Now, by considering Half-Sommerfeld Boundary Conditions and solving Eq. (6)

for p̄ it is possible to compute the pressure field in dimensionless form. This
expression is known as the Ocvirk solution and is defined by [30,31]:
 2
L ε sin θ 
P ocv = −6π 1 − z2 (7)
D 1 + ε cos θ

On the other hand, the mathematical model for the pressurized case employs
an oil injection port. Thusly, to simulate the effect of the external pressuriza-
tion in the hydrodynamic journal bearing
 the generalized spatial
 Dirac delta
∞ θ − θin ∞ z−a
function is used. Thus, δ = [θ − θin ] ; δ = [z − a] where,
0 θ = θin 0 z = a
θin = π + β − φ. The symbolic function δ(z − a) represents the pressure in
a source point located in z = a, and the accumulated pressure is termed as
the Heaviside function [32], in such a way ΔFpres = Δp · Δs, where, ΔFpres
is the pressurization force, [N ]; Δs is the area of the oil injection port, [m2 ],
and Δp is the pressure of the oil injection port, [P a]. Consequently, when
Δp → 0, then Δs → 0 so that: ΔFpres = Δp · Δs = q = constant. These
conditions provide the mathematical model that represents a pressurization
point in the dimensionless form described by the spatial Dirac delta func-
tion, which results in: (Δp)prt = q prt δ(z − a)δ [θ − (π + β − φ)] [33]. Hence,
δF δF f
S ; Pdim = µN Cr . Knowing that, (Δp)prt is the
pres pres R
q̄prt = DLP dim
= Fdim = prt
dimensionless pressure in the oil source port; q prt is the dimensionless pressure
in the point source regard to the equivalent characteristic force; fprt is dimen-
sionless pressurization force in the point source respect to the weight; S is the
dimensionless Sommerfeld’s number, δ(z) is the Dirac delta function; a is the
arbitrary dimensionless axial position for the oil injection port; β defines the
 Attenuation of the Amplitude Vibration Level 185

port position or angular coordinate of the oil injection port, [rad]; and Fdim is
the equivalent characteristic force, [N ]. Then, the solution of Eq. (6) for p̄pres
becomes:
 2
L δ [θ − (π + β − φ)]
p̄pres (z̄, θ) = q̄prt (1 − az̄ − |(z̄ − a)|) (8)
D 2(1 + cos θ)3

where, n̄ is the peak value of Dirac Delta approximation. Finally, the total pres-
sure field for hybrid short journal bearing with a simple oil injection port is given
by the addition of the unpressurized mathematical model (classic case) Eq. (7)
plus the pressurized solution (pressurized case) Eq. (8), which can be written as:
 2
L ε sin θ 
p = pocv + ppres = −6π 1 − z2
D 1 + ε cos θ
 2
L δ [θ − (π + β − φ)]
+ q prt [1 − az − | (z − a) |] (9)
D 2(1 + ε cos θ)3

Two particular cases of study are analyzed: first, two horizontal oil injection
ports located at β = π/2 and β = 3π/2, and second, at the top β = π and
bottom β = 2π, respectively. The peak value of the Dirac Delta approximation
is proposed to simulate an adequate function of the equivalent oil injection port.
The equilibrium eccentricity ε = e/Cr is evaluated from ε = 0.1 to 0.9 [10] and
is located in the middle of the journal bearing respect to the Z-axis to suppress
conditions of imbalance forces. Plotting Eq. (9) and considering ñ = 10, ε = e/Cr
from 0.01 to 0.9, a = 0 and fprt from 0 to 100 results:

2 2

1.5 1.5

1 1
p̄

p̄

0.5 0.5

0 0
1 1
2*π 2*π
0 3*π/2 0 3*π/2
π π
π/2 π/2
z̄ -1 0
θ z̄ -1 0
θ

(a) (b)

Fig. 3. Results for the computer simulation where: (a) a = 0, fprt = 20, β = 3π/2,
 = 0.5, S = 1.6, φ = 0.99 (b) a = 0, fprt = 20, β = π/2,  = 0.5, S = 1.8, φ = 0.56.
186 D.-E. Rivera-Arreola et al.

2 2

1.5 1.5

1 1
p̄

p̄
0.5 0.5

0 0
1 1
2*π 2*π
0 3*π/2 0 3*π/2
π π
π/2 π/2
z̄ -1 0
θ z̄ -1 0
θ

(a) (b)

Fig. 4. Results for the computer simulation where: (a) a = 0, fprt = 20, β = π,
 = 0.48, S = 2, φ = 0.96 (b) a = 0, fprt = 20, β = 2π,  = 0.39, S = 2, φ = 1.07.

Figs. 3 and 4 illustrate the hydrodynamic pressure in the oil film on the sliding
surface of a hybrid short journal bearing by a specific port respectively.

3 PID Control Techniques

A nonlinear dynamic system can usually be represented by a set of nonlinear
differential equations in the form: ẋ = f (x, u, t) and y = h (x, u, t). To obtain a
full description of the system for further analysis, the following state variables
are used: x1 = θx , x2 = θy , x3 = θ̇x , x4 = θ̇y . In such a way to determine the
stability of the nonlinear system Eqs. (2) and (3) and thus ensure the optimal
functionality of the PID controller, the Lyapunov’s method for stability is used.
First, the direct method of stability criterion states that: be the origin x = 0 in
an equilibrium point of ẋ = f (x) and be D ⊂ n a domain that contains the
origin. Be V →  a continuously differentiable function such that: V (0) = 0 and
V (x) ≥ 0 in D − {0}. Thus, V̇ (x) ≤ 0. In D, then x = 0 the nonlinear rotor-
system is asymptotically stable [34–36]. In this sense, the proposed Lyapunov's
equation based on the mathematical model of the energy is:
1 1 1 1
V (x) = kt x21 + kt x22 + IO x23 + IO x24 (10)
2 2 2 2
Satisfying that V (x) ≥ 0 where: kt = GJl P is the torsional stiffness, [N m]; G
is the shear modulus of the shaft, [P a]; JP is the polar moment of inertia [m4 ].
Thus,
∂V (x) ∂V (x) ∂V (x) ∂V (x)
V̇ (x) = ẋ1 + ẋ2 + ẋ3 + ẋ4 (11)
∂x1 ∂x2 ∂x3 ∂x4
The computed results can define that the nonlinear rotor-system is stable in
the sense of Lyapunov where, 0 ≤ x1 ≤ 71µm, 0 ≤ x2 ≤ 71µm, 0 ≤
x3 ≤ 6000 rpm., 0 ≤ x4 ≤ 6000 rpm.. To validate the proposed Eq. (10)
the indirect method of Lyapuno’s stability was also used. This approach uses the
 Attenuation of the Amplitude Vibration Level 187

linearization of the system by means of the Jacobian matrix for determining the
local stability of the original system. Thus, the stability criterion states that:
be the origin x = 0 an equilibrium point of the nonlinear system ẋ = f (x)
where f : D ⊂ n is a continuously differentiable function and D ⊂ n is an
environment of the origin. Hence, consider the system: ẋ = f (x, t) with f =
(x,t) 
(0, t) = 0 for all t ≥ 0. Define by A (t) = ∂f∂x  . To be the Jacobian matrix
x=0
of f (x, t) regarding to x, evaluated. Now, considering a quadratic Lyapunov
function candidate [37] V (x) = xT P x, where P is a real symmetric positive
definite matrix. The derivative of V along the trajectories of the linear system
Eq. (12) is given by:
V̇ (x) = xT P ẋ + ẋT P x = −xT Qx ≤ 0 (12)
where Q is a symmetric matrix defined. If Q is positive definite, we can conclude
that the origin is asymptotically stable; that is, Reλi < 0 for all eigenvalues of
A, in other words must be positive det (P − λI) ≥ 0. Currently, PID controllers
are distinguished by their flexibility, reliability and computational advantage.
From this point of view, active control was applied to the proposed rotor-system
configuration to reduce the amplitude of the shaft vibration crossing its first
resonance. Thus, if controlled actively, Eqs. (2) and (3) becomes [38]:
ẋ1 = x3
ẋ2 = x4
9l2 cxx x3 (1+tg 2 x1 ) 9l2 cxy x4 (1+tg 2 x2 ) 9l2 ukxx tan x1
ẋ3 = − 16IO − 16IO − 16IO
9l2 ukxy tan x2 (13)
− 16IO + M1
IO + u
IO
9l2 cyx x3 (1+tg 2 x1 ) 9l2 cyy x4 (1+tg 2 x2 ) 9l2 ukyx tan x1
ẋ4 = − 16IO − 16IO − 16IO
9l2 ukyy tan x2
− 16IO + M2
IO + u
IO

where:
l  
M1 = md1 ω 2 ex1 cos ωt − αey1 sin ωt + lmd2 ω 2 ex2 cos ωt − αey2 sin ωt (14)
2
l  
M2 = md1 ω 2 ex1 sin ωt + αey1 cos ωt + lmd2 ω 2 ex2 sin ωt + αey2 cos ωt (15)
2
besides, u(t) is the active control represented by the characteristic PID and can
t
be expressed mathematically as: u(t) = Kd ė (t) + Ki e (t) dt + Kp e (t), where:
0
Kp , Ki , and Kd are the coefficients for the proportional, integral, and derivative
terms respectively and denote tuning parameters. Also: e = XR − X1 is the
error, where XR is the setpoint and X1 is the process variable. The classic tuning
Ziegler-Nichols (ZN) method was selected due to its efficiency and fast response.
K
Therefore, the tuning parameters were calculated as: Ki = Tip , Kd = Kp Td and
Ku = 4 (h − (DB/2)) /πd, knowing that (ZN) Kp = 0.60 ∗ Ku , Ti = 0.50 ∗ Tu
and TD = 0.125 ∗ Tu where, h is the amplitude of the relay, d is the amplitude
output of the plant and Tu is the ultimate period of the closed-loop system in
the steady-state of the test [24,39,40].
188 D.-E. Rivera-Arreola et al.

3.1 Simulation

To validate the numeric simulation, the following parameters presented in the

Table 1 were taken from the experimental test. The simulation results were per-
formed in Matlab-Simulink R
using ODE45. Hence, by simulating Eqs. (2) and
(3) the Fig. 5 depicts for (a), (b) and (c) the vibratory and orbit response for
β = π without control techniques and its correspondence with the experimental
test in Fig. 7. However when the simulation is performed the Eq. (13) depicts the
vibratory and orbit response for β = π by using active control techniques pre-
sented in Fig. 5 for (d), (e) and (f). The behavior was smooth and the amplitude
did not exceed the response at 5µm (peak).

10-5 Vibratory response x 10-5 Vibratory response y

4 4

2 2
Amplitude [m]

Amplitude [m]

0 0

-2 -2

-4 -4
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Frequency [rpm] Frequency [rpm]

(a) (b)

10-5 Orbit of the shaft centerline 10-5 Orbit of the shaft centerline

2
Amplitude axis Y [m]

Amplitude axis Y [m]

1
0
0

-1 -1

-3 -2 -1 0 1 2 3 4 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2

a) Amplitude axis X [m] 10-5 Amplitude axis X [m] 10-5

(c) (d)

10-5 Vibratory response x 10-5 Vibratory response y

4 4

2 2
Amplitude [m]

Amplitude [m]

0 0

-2 -2

-4 -4
0 1000 2000 3000 4000 5000 6000 0 1000 2000 3000 4000 5000 6000
Frequency [rpm] Frequency [rpm]

(e) (f )

Fig. 5. System response as a function of the rotational frequency where: without law
control (a), (b) and (c), and with active control techniques (d), (e) and (f) from ω = 200
to 6000 rpm., ωn = 4550 rpm and β = π.
 Attenuation of the Amplitude Vibration Level 189

4 Experimental Test
Fig. 6 illustrates the platform Bently Nevada R
model RK4 rotor kit [41]. In the
experimental setup “H” is the electrical motor a.c. that has a speed controller
and acceleration ramp, “O” is a pivot self-aligning bronze bushing, “C” is the
hydrodynamic journal bearing which can be pressurized by means of a pump;
“A” and “B” are the positions of the equidistant disks in the rotor-system. The
second disk is not visible since it is inside the supporting shell B. The parameters
for the overall system are given in Table 1.

Fig. 6. (a) Modified Rotor Kit type Bently Nevada

R
and (b) Graphical representation.

Table 1. System Parameters

Definition Symbol Value SI units

Acceleration due to gravity g 9.81 [m/s2 ]
Angular speed considering counterclockwise rotation ω 0 − 6000 [rpm.]
Angular acceleration α 0, 1.2 [rad/s2 ]
Damping factor ζ 0.45 [dimensionless]
Diameter of the journal bearing D 0.0254 [m]
Disk mass1 md1 0.8 [kg]
Disk mass2 md2 0.8 [kg]
Distance of the residual unbalance of mass ex x
1 = e2 0.7 × 10−6 [m]
Distance of the residual unbalance of mass ey y
1 = e2 0.7 × 10−6 [m]
Length of the journal bearing L 0.0127 [m]
Length of the shaft l 0.24 [m]
Lubricant viscosity µ 0.0124 [Pa.s]
Polar moment of inertia Jp 4.0863 × 10−8 [m4 ]
Radial clearance Cr 71 × 10−6 [m]
Radio of both Disks rd1 = rd2 0.0367 [m]
Shear modulus of the shaft G 77.2 × 109 [Pa]
Weight of the system W = Σmd · g 15.6960 [N]
190 D.-E. Rivera-Arreola et al.

Fig. 7. Experimental test with external oil supply without law control considering
β = π, ωn = 4600 rpm where: (a) Experimental bode diagram for X-axis and (b)
Experimental bode diagram for Y -axis.

5 Conclusions

In this paper, a simple mathematical model for a rigid rotor-system with two
disks supported by a single hybrid short journal bearing has been proposed and
analyzed. The analytical expressions were derived by arranging four oil injection
points orthogonally and modeled employing the Dirac Delta function. The equa-
tions yield results that support the experimental tests. These test successfully
validate the applied lubricant pressurizing action; yet, currently the experimen-
tal set up lacks rapid active control capability. Also, to enhance the rotor-system
stability with a wider and safer operational frequency range a PID active control
technique was applied to the numerical method. To analyze the stability of the
nonlinear rotor-system two different approaches based on the methodology of
the direct and indirect Lyapunov’s theory were used. These mathematical tools
allowed to ensure the optimal functionality of the PID controller. The results
have shown that the PID active control technique led to reducing significantly
the amplitude of the shaft vibration crossing its first resonance. Moreover, the
proposed experimental system shows that by increasing the bearing’s pressure
supply at β = π and β = 3π/2 the amplitude vibration level can be attenuated.
Finally, the results of the numerical solutions and tests show that the simplicity
of the proposed rotor-system can be useful for designing and analyze new options
of control laws.

References
1. Muszynsk A (1986) Whirl and whip-rotor bearing stability problems. J Sound Vib
111:443–462. https://doi.org/10.10007/1234567890
2. Tuma J, Bilos J (2007) Fluid induced instability of rotor systems with journal
bearings. Eng Mech 14:69–80
3. Martinez E, Nossov VR, Gómez-Mancilla JC (2004) Validation and analysis of
dynamic operation of a rotor-bearing system with hybrid lubrication. J Brazilian
Soc Mech Sci Eng 7:83–90
 Attenuation of the Amplitude Vibration Level 191

4. Garg HC, Sharda HB, Kummar V (2006) On the design and development of hybrid
journal berings: a review. Tribotest 12:1–19. https://doi.org/10.1002/tt.1
5. Donald BE (2002) The rotor system model. Fundamentals of rotating machinery
diagnostics, pp 499–515
6. Fedor JV (1961) Journal Bearings With Arbitrary Position of Source. J Basic Eng
ASME 83:572–578. https://doi.org/10.1115/1.3662269
7. Lund JW (1979) Evaluation of stiffness and damping coefficients for fluid-film bear-
ings. Shock Vib Digest 11(1):5–10. https://doi.org/10.1177/058310247901100103
8. Childs DW (1993) Rotordynamic introduction to hydrodynamic bearing and
squeeze-film Dampers. Turbomachinery Rotordynamics: Phenomena, Modeling
and Analysis, pp 132–223
9. Norton RL (2006) Hydrodynamic lubrication theory. Machine Design: An Inte-
grated Approach, USA, pp 653–664
10. Michael MK, Richard BE (2008) Reynolds Equation and Applications. Applied
tribology: bearing design and lubrication, USA, pp 143–171
11. Andras ZS (2005) Thick-Film Lubrication. Fluid Film Lubrication, pp 88–146
12. Chasalebris A, Sfyris D (2013) Evaluation of the finite journal bearing character-
istics, using the exact analytical solution of the Reynolds equation. Tribology Int
57:216–234. https://doi.org/10.1016/j.triboint.2012.08.011
13. Woods CM, Brewe DE (1989) The solution of the elrod algorithm for a dynamically
loaded journal bearing Using multigrid techniques. Trans ASME J Tribol 111:302–
308. https://doi.org/10.1115/1.3261914
14. Goenka PK (1984) Dynamically loaded journal bearings: finite element method
analysis. Trans ASME J Tribol 106:429–439. https://doi.org/10.1115/1.3260954
15. Gómez-Mancilla JC, Nossov VR, Silva-Navarro G (2005) Rotor bearing system
stability performance comparing hybrid versus conventional bearings. Int J Rot
Mach 16–22:2005. https://doi.org/10.1155/IJRM.2005.16
16. Booker JF (1965) Dynamically loaded journal bearings: mobility method of solu-
tion. ASME J Basic Eng 3:537–546. https://doi.org/10.1115/1.3650602
17. Childs D, Moes H, Leeuwen H (1976) Journal Bearing Impedance Description for
Rotordynamic Applications. Trans ASME J Lub Tech 1–24 https://doi.org/10.
1115/1.3453021
18. Moes H, Sikkes EG, Bosma R (1986) Mobility and impedance tensor methods for
full and partial-arc journal bearings. Trans ASME J Tribol 108:612–620. https://
doi.org/10.1115/1.3261282
19. Nicoletti R, Santos IF (2002) Linear and non-linear control techniques applied to
actively lubricated journal bearings. J Sound Vib 260:927–947. https://doi.org/10.
1016/S0022-460X(02)00951-3
20. Santos IF, Watanabe FY (2003) Feasibility of influencing the dynamic fluid film
coefficients of a multirecess journal bearing by means of active hybrid lubri-
cation. J Brazil Soc Mech Sci Eng 25:154–163. https://doi.org/10.1590/S1678-
58782003000200007
21. Alejandro CV, Santos IF (2014) Tilting-pad journal bearings with active lubrica-
tion applied as calibrated shakers: theory and experiment. J Vib Accoust 136:1–11.
https://doi.org/10.1115/1.4028452
22. Arias-Montiel M, Silva-Navarro G, Antonio-Garcia A (2014) Active vibration con-
trol in a rotor system by an active suspension with linear actuators. J Appl Res
Technol 12:898–907. https://doi.org/10.1016/S1665-6423(14)70596-6
23. Jorge GS, Santos IF (2017) Active tilting-pad journal bearings supporting flexible
rotors: Part I –The hybrid lubrication. Tribo Int 107:94–105. https://doi.org/10.
1016/j.triboint.2016.11.018
192 D.-E. Rivera-Arreola et al.

24. Leonardo BSM, Diogo SA, Ricardo UM, Katia LC (2019) Reduction of rotor
vibration amplitude using PID tuning methods. IFToMM 2018. Mechanisms and
Machine Science, vol 62, pp 74–88 https://doi.org/10.1007/978-3-319-99270-9 6
25. Gómez-Mancilla JC, Meda CJA, Torres CSG, Nossov VR (2005) Model deriva-
tion and numerical simulation for a pivoted rigid rotor-bearing system. Key Eng
Mater 15:89–94. https://doi.org/10.4028/www.scientific.net/AMM.15.89 Trans
Tech Publications ltd
26. Nosov VR, Gómez-Mancilla JC, Rivera-Arreola DE, Ramı́rez VI (2010) Stationary
position for hybrid journal bearing with horizontal Point injection ports. Key Eng
Mater 449:29–36. https://doi.org/10.4028/www.scientific.net/KEM.449.29 Trans
Tech Publications ltd
27. Nosov VR, Rivera-Arreola DE, Gómez-Mancilla JC (2013) Stability threshold
velocity for a hybrid short journal bearing with the injection ports in the sym-
metric points. J Sci Res Rep 2:74–92. https://doi.org/10.9734/JSRR/2013/2142
28. Nosov VR, Gómez-Mancilla JC (2010) On the appearance of lubricant film rupture
in cylindrical journal bearings. Tribology Trans 47:233–238. https://doi.org/10.
1080/05698190490439120
29. Nosov VR, Gómez-Mancilla JC, Meda CJA (2011) Occurrence of hydrodynamic
cavitation. Water Sci Technol 64:595–601. https://doi.org/10.2166/wst.2011.608
30. Ocvirk FV (1952) Short-bearing approximation for full journal bearings. NASA
Tech Rep Serv (NTRS) 2808:1–61
31. Stachowiak GW, Andrew AW (1993) In hydrodynamic lubrication. Engineering
tribology, pp 121–206
32. Kanwal RP (1998) The Dirac delta function and delta sequences. Generalized
Functions: Theory and Applications, pp 1–14
33. George A, Hans W, Frank HE (2012) Dirac delta function. Mathematical Methods
for Physicists, pp 83–95
34. Lyapunov AM (1992) The general problem of the stability of motion, pp 19–247
35. Afanasev VN, Kolmanovskii VB, Nosov VR (1995) Mathematical theory of control
system design. 3:–651
36. Jean-Jacques E, Slotine E, Weiping L (1991) Applied nonlinear control, pp 1–433
37. Khalil HK (1996) Lyapunov Stability. Nonlinear Systems, 2nd edn, pp 120–128
38. Tammi k (2007) Active Control of Radial Rotor Vibrations: Identification, Feed-
back, Feedforward, and Repetitive Control Methods. Doctoral thesis: Department
of Automation and Systems Technology at the Helsinki University of Technology,
Espoo, Finland, pp 1–165
39. Ziegler JG, Nichols NB (1943) Optimum settings for automatic controllers. Trans
ASME 65:433–444
40. Parinya A, Zongli L (2013) An experimental study on PID tuning methods for
active magnetic bearing systems. Int J Adv Mech syst 5(2):146–154. https://doi.
org/10.1504/IJAMECHS.2013.055991
41. SEPILAB Instituto Politécnico Nacional, Laboratorio de vibraciones y
rotodinámica. Ciudad de México, México
 Transient Vibrations Suppression
in Parametrically Excited Resonators

Miguel Ramı́rez-Barrios1(B) , Fadi Dohnal2 , and Joaquı́n Collado3

1
Professional Interdisciplinary Unit of Biotechnology,
National Polytechnic Institute, Mexico City, Mexico
mlramirezb@ipn.mx
2
Division for Mechatronics Lienz, UMIT - Private University for Health Sciences,
Medical Informatics and Technology, Lienz, Austria
fadi.dohnal@umit.at
3
Automatic Control Department, Cinvestav, Mexico City, Mexico
jcollado@ctrl.cinvestav.mx

Abstract. Micromechanical resonators with a high-quality factor have

been studied in recent years due to their excellent performance
and potential applications. Some of these micromechanical resonators
describe a long settling time, which can be disabled for fast switching
applications. A novel design reduces the settling time by coupling two
micromechanical resonators with parametric excitation; the parametric
excitation tuned at the anti-resonant frequency increases the damping
into the system; thus, the responses vibrations are attenuated, and the
settling time is minimized. The scheme can be classified as open-loop
control and belongs to the vibrational control field. Through a mathe-
matical approach, this article describes this phenomenon in detail; by
the averaging technique, the parametric excitation effects are analyzed.
Moreover, we give new analytic conditions and formulas to minimize
the settling time; these formulas might be helpful for future experimen-
tation. Finally, using the current analysis, this contribution proposes a
methodology to suppress the transient vibrations in this class of systems.

Keywords: Parametric excitation · Vibrational control ·

Micromechanical resonators

1 Introduction

Due to their high-quality factor (Q = ωi /γi ) and accurate resonance spectra,

high-Q mechanical resonators have been studied and used in many applications,
such as filters, high precision sensors, amplifiers, and logic circuits [10]. However,

This project was partially funded by the Austrian IWB-project:LaZu-COLD IWB

TIROL POST 005 (Campus Technik Lienz).
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 193–205, 2020.
https://doi.org/10.1007/978-3-030-45402-9_19
194 M. Ramı́rez-Barrios et al.

Fig. 1. Weakly coupled resonators scheme. The time response of beam one describes a
long settling time ts .

high-Q resonators have a long settling time, i.e., its transition time to steady-
state is slow, which is challenging for rapid repetitive operations in this class of
devices. Recently, an experimental implementation demonstrated for the settling
time reduction using a piezoelectrically induced parametric in paired microme-
chanical resonators [10,11]. These paired micromechanical resonators can be
represented in a scheme, as it is shown in Fig. 1, Fig. 2 and Fig. 3. The experi-
mentation, reported in [10,11], proves that a periodic signal adequately tuned,
can be used as an open-loop control to minimize the response’s vibrations in
one mode of the system. This experiment describes how to use the parametric
excitation to fast transfer energy between the two beams of the system. With
this idea, the parametric excitation is activating during a short time window,
called a pulse, to transfer the response’s energy in the first mode into the second
mode. Thus, the second mode dissipates the energy of the system, wherewith,
the transient vibrations on the first mode are suppressed; therefore, the settling
time in the first mode is minimized. The above experiment can be classified as
a particular case of the anti-resonance phenomenon [5]. In the current contri-
bution, we explain in a mathematical point of view these experiments; then, we
provide an analytical formula to compute the length of the pulse.
The dynamics of motion of two micromechanical resonators coupled by para-
metric excitation can be expressed by two coupled Mathieu equations: [11]:

ẍ1 + γ1 ẋ1 + ω12 x1 + β cos (ωp t) x2 = f1 cos (ωd t)

(1)
ẍ2 + γ2 ẋ2 + ω22 x2 + β cos (ωp t) x1 = f2 cos (ωd t)

where γi is the constant damping factor, β is the amplitude of the parametric

excitation signal p(t) = βcos(ωp t), and fi cos(ωd t) is the drive signal (or forced
signal). Many publications deal with single or coupled differential equations with
periodic coefficients, where it is possible to find analytic and numerical methods
to describe the dynamics behavior. The main focus of these works is the desta-
bilizing effect of parametric excitation [2,3,19]. Other publications [1,16] deal
with the stabilizing effect of the parametric excitation.
 Transient Vibrations Suppression in Parametrically Excited Resonators 195

Fig. 2. Strongly coupled resonators scheme, the coupling term is the parametric exci-
tation tuned at the anti-resonance frequency ωp = ω2 − ω1 .

On the other hand, the suppression of vibrations by parametric excitation

[5,7,12,13] arises from the experiments reported in [17]. This experiment
establishes that in the system of Eq. (1), the parametric excitation effect
increases the damping in the system when the parametric excitation is tuned
at the specific frequencies: ωp = |ω2 −ω
n
1|
with n ∈ N+ , these are called as anti-
resonance frequencies. By singular perturbations and averaging techniques the
treatment [6] studies and describes the anti-resonance phenomena in detail.
Since the linear time-varying system, Eq. (1), cannot be solved analytically,
asymptotic methods become powerful analytical tools. Well-known techniques
such as the Poincaré Lindstedt method, harmonic balance, averaging method,
multiple scales method, and successive approximations assume that β is a small
parameter; thus the above techniques provide valid approximations on the time-
scale 1/β. Mathematicians, physicists, and engineers have used the techniques
listed above because they allow computing analytic formulas that describe the
system dynamics [8,9]. A numerical method based on the Lyapunov-Floquet
theory can be found in [3,14,15], which allows arbitrary values of β. Therefore,
to describe the experiments presented in [11], this article deals with the averaging
method developed in [6,8]. We analyzed the dynamics behavior when the system
is parametrically excited at anti-resonance frequency ωp = ω2 − ω1 . Averaging
the Eq. (1), the linear time-variant system is transformed into an approximated
linear time-invariant represented by the amplitude-phase variables. With this
theory, we obtain an analytic condition to guarantee the transfer of the energy
between the two modes of the system. Then we provide a formula to compute
the length of the pulse of parametric excitation. This analysis establishes an
algorithm to suppress transient vibrations on micromechanical resonators.
The paper is organized as follows: The next section establishes the system
description and the problem to solve. Then, Sect. 3 briefly describes the aver-
aging method for the Eq. (1). The following section analyzes the effect of the
196 M. Ramı́rez-Barrios et al.

Fig. 3. Strongly coupled resonators scheme, the coupling term is a pulse of parametric
excitation. The settling time of beam one ts is minimized.

anti-parametric resonance and gives the formula to compute the length of

the pulse which guarantees the suppression of vibration in the first beam of
the system. In the last section, we reproduce the experiments of [11] to confirm
the successful prediction of the formulas.

2 Problem Statement

The Fig. 1 depicts a schematic representation of two coupled microelectrome-

chanical resonators, where it is assumed that each beam resonates at ω1 and
ω2 respectively. Same Fig. 1 shows the time response of each beam, defined x1
and x2 in Eq. (1). Because ωd = ω1 , the beam one resonates1 with drive sig-
nal f1 cos(ωd t) during the time td . The free response of the system starts when
f1 = 0, this reveals settling time ts in the first beam. Thus, this class of devices
demands the waiting for at least ts to initialize again, which disabling them to
quickly operations.
If the two mechanical resonators are coupled by parametric excitation prop-
erly tuned, the vibration energy in mode one (beam 1) can be quickly transferred
to the second mode (beam 2), as shown in the Fig. 2. Here, the envelope ampli-
tudes of the responses describe an anti-phase behavior. Then, if the parametric
excitation is turning off at the time tp , the vibration energy in the first mode
is fast transferred to the second mode during tp , after the time tp the second
mode dissipates the energy vibration of the system, leading to the significant
reduction of the settling time ts in the first mode, see the Fig. 3. We named

1
This fact can be understood, when the micromechanical resonator is used as a sensor.
 Transient Vibrations Suppression in Parametrically Excited Resonators 197

the parametric excitation signal active in the time tp as a pulse of parametric

excitation. Namely, with a pulse of parametric excitation, the transient vibra-
tions of the beam one are suppressed. The present contribution fully describes
this phenomenon in a mathematical sense, which leads to obtaining a formula
to compute tp .

3 Transformation and Averaging

The experiments with new models of high-Q micromechanical resonators [11]
can be dynamically described by the system Eq. (1). For the sake of simplicity,
we applied the averaging technique to the homogeneous part of the system only.
Then, the system can be rewritten as:

ẍ + Γ ẋ + (Ω + Bp (t)) x = 0, (2)
 T
with x = x1 x2 , where p(t) = β cos (ωp t) is the parametric excitation signal,
being the matrices
   2   
γ 0 ω 0 b b
Γ = 1 , Ω = 1 2 , B = 11 12 .
0 γ2 0 ω2 b21 b22

Rewritten in the comprehensive index notation (Einstein summation,

repeated indices are implicitly summed over):

ẍi + ωi2 xi = −β {γi ẋj − bij xj cos (ωp t)} (3)

with i, j = 1, 2 and applying the time transformation

τ = ωp t (4)

to normalize the frequency, the Eq. (3) turns to

β  

xi + νi2 xi = − ω γ x
p i j + b x
ij j cos (τ ) (5)
ωp2

d
where dτ () = () and νi = ωωpi . To study the behavior of (5) one assumes that the
solutions can be written into the amplitude Ai (τ ) and phase φi (τ ) terms [18]:

xi (τ ) = Ai (τ ) ci + φi (τ ) si (6)

and
xi (τ ) = −Ai (τ ) νi si + φi (τ ) νi ci (7)
to simplify the process, we omitted the time dependence A = A (τ ), φ = φ (τ )
and with the abbreviations si = sin (νi τ ), ci = cos (νi τ ), we performed the
coordinate transformation (6)–(7) into (5):

Ai ci + φi si = 0 (8)

198 M. Ramı́rez-Barrios et al.

by differentiating (7) the next relation holds

xi = −Ai νi si − Ai νi2 ci + φi νi ci − φi νi2 si

and substituting into (5) we obtain

β
− Ai νi si + φi νi ci = F̃i (9)
ωp2

where F̃i = F̃i (A1 , A2 , φ1 , φ2 , τ ) is given by:

⎧ ⎫
⎨ ⎬
F̃i = − ωp γi [(−Aj νj ) sj + φj νj cj ] + bij (Aj cj + φj sj ) cos (τ )
⎩ ⎭
j j

 T
if we define the state vector u = A1 A2 φ1 φ2 and insert the relation (8) in
(9) the equations of motion finally become:

Ai = − ωpβωi F̃i (u, t) si = β G̃i (u, τ )

(10)
φi = ωpβωi F̃i (u, t) ci = β H̃i (u, τ )

with i = 1, 2. The functions G̃i (u, τ ) and H̃i (u, τ ) on the right-hand side of
these system of equations are not periodic, they are quasi-periodic. However,
they can be split into a finite sum of different periodic terms:
N N
G̃i (u, τ ) = G̃i,k (u, τ ) and H̃i (u, τ ) = H̃i,k (u, τ )
k=1 k=1

where N is fixed, G̃i,k (u, τ ) and H̃i,k (u, τ ) are Tk periodic functions in τ . In
this case, the averaging in the general case method can be applied [4,18]:


N  Tk
Âi = β 1
Tk 0
G̃i,k (u, τ ) dτ
k=1
 Tk (11)
N
φ̂i = β 1
Tk 0
H̃i,k (u, τ ) dτ
k=1

where (ˆ) indicates time-averages variables. Hence, for averaging the system,
the periods of the right hand sides of Eq. (10) must be determined. Using the
trigonometric identities, products of the trigonometric terms on the right hand
sides of Eq. (10) can be rewritten as a sum of basic trigonometric periodic terms.
Thus, the integrals in Eq. (11) over a basic trigonometric terms yield always zero,
|ω2 ±ω1 |
with the exception of the frequencies ωp = 2ω i
n and ωp = n with n ∈ N+ ,
namely resonant frequencies [18,19].
 Transient Vibrations Suppression in Parametrically Excited Resonators 199

4 Parametric Anti-resonance Frequency

Averaging the Eq. (10) at the anti-resonance frequency: ωp = ω2 − ω1 we obtain
the averaged system:
û = Aû (12)
with ⎡ ⎤
− 12 γ1 0 0 1
4ω1 b12
β ⎢ 0 − 2 γ1 − 4ω1 b12 0 ⎥
1 1
A= ⎢ ⎣
⎥
ωp 0 4ω2 b21 − 2 γ2
1 1
0 ⎦
− 4ω1 2 b21 0 0 − 21 γ2
a linear time invariant system, where the difference between the original u =
 T  T
A1 A2 φ1 φ2 and the averaged solutions û = Â1 Â2 φ̂1 φ̂2 is close to β
order, û (τ ) − u (τ ) = O (β) on the time scale 1/β [18].

x Response x Response
Amplitude envelope Amplitude envelope

t t
2
(a) (γ1 − γ2 ) > b12 b21
ω1 ω2
>0 (b) b12 b21
ω1 ω2
> (γ1 − γ2 )2 ≥ 0

Fig. 4. Amplitude envelope conditions. a) Exponential envelope time-response b) Beat-

ing envelope time-response.

The motion system (12) does not depend of terms b11 and b22 , i.e., the cou-
pling terms b12 , b21 fully describe the effect of the parametric excitation at
ωp = ω2 − ω1 . Then computing the eigenvalues of the matrix A, we obtain
  
β 2 b12 b21
λ12 = − (γ1 + γ2 ) ± (γ1 − γ2 ) − (13)
4ωp ω1 ω2

(the matrix has two repeated eigenvalues: λ1 = λ3 and λ2 = λ4 given by relation

(13)). The transformation of the system to the phase amplitude coordinates
200 M. Ramı́rez-Barrios et al.

provides the ‘slow motion’ which is represented by the amplitude envelope time
evolution response, see Fig. 4. Then the response envelope can be described by
the eigenvalues λi of the matrix A. If the condition

2 b12 b21
(γ1 − γ2 ) > ≥0 (14)
ω1 ω2
is satisfied the eigenvalues are real. Therefore, the envelope decreases or increase
exponentially as is shown in the Fig. 4(a).
Alternatively, when the matrix A has complex eigenvalues the parametric
excitation term introduces a modulation in the slow motion, thus if the inequality
b12 b21 2
> (γ1 − γ2 ) ≥ 0 (15)
ω1 ω2
is fulfilled the parametric excitation leads to a modulated responses signals,
sometimes called as a ‘beating signal’. The imaginary part of λi can be inter-
preted as: the frequency of the slow motion. Thus, the frequency of the
envelope amplitudes is given by:

β b12 b21 2
ω̂A = − (γ1 − γ2 ) (16)
4ωp ω1 ω2

thus, the envelope amplitude responses of Eq. (12) describe a ‘beating’ behavior,
see the Fig. 4. Furthermore, there are two amplitude responses, since there are
two frequencies ±ω̂A the two envelope responses are in anti-phase when ω =
ω2 − ω1 and Eq. (15) is fulfilled. Therefore, the envelope amplitude in the time
response in Eq. (1) can be approximated with the averaged system in Eq. (12).
Rescaling the time τ → t by the Eq. (4) the period TA of envelope amplitude
of the responses in Eq. (1), can be computed as:
8π
TA =  (17)
2
β b12 b21
ω1 ω2 − (γ1 − γ2 )
 Transient Vibrations Suppression in Parametrically Excited Resonators 201

5 Simulations and Results

Table 1. Simulations parameters obtained from the experiments in [11].

Natural frequencies Parametric Anti-Resonance

ω1 = 291.51 × 2π × 103 rad/s ωp = ω2 − ω1 = 440 × 2π rad/s
ω2 = 291.95 × 2π × 103 rad/s β=1V
Coupling Coefficients Dissipation Coefficients
b = b12 = b21 = 1.2 × 109 and b11 = b22 = 0 γ1 = γ2 = 39
Drive signal Quality factor
⎧
⎨ 0.9 V 0 ≤ t ≤ 14 ms
⎪
ω1
ωd = ω1 , f1 = f2 = 0.4 V 14 ms< t ≤ 0.1 s Q= = 46964
⎪
⎩
γ1
0 t > 0.1 s

Simulating the system (1) for the experiment parameters in [11], which are shown
in the Table. 1. The Fig. 5, Fig. 6 and Fig. 7 show the expected behavior in the
averaged method. The three figures depict the time responses for beam one and
two, in blue and red color, respectively. The drive signal is active during the

10-12
5
x
2.5 x
volts

-2.5

-5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
-3
time (s)
10
1
Drive signal f 1
volts

-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)
1
volts

0
Parametric excitation signal =0 volts
-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)

Fig. 5. Simulation for parameters in Table 1. No parametric excitation. The time

response in beam one exhibits a long settling time.
202 M. Ramı́rez-Barrios et al.

time td = 0.1 s, in green color. And the parametric excitation signal in brown
color. The Fig. 5 shows the natural response of the system without parametric
excitation, after the time tp = 0.1, the time response of beam one exhibits a long
settling time approximately ts = 0.1 s.
Since the mathematical model proposed in [11] considers: b11 = b22 = 0,
b = b12 = b21 and γ = γ1 = γ2 , the condition (15) is always fulfilled and the
envelope amplitude period is given by the formula (17), which is simplified to
√
8π ω1 ω2
TA = . (18)
βb
Then, in Fig. 6 and also in Fig. 8 can be observed the expected anti-phase
behavior in the envelope amplitude time-responses, then using the Eq. (18) the
period of the envelope amplitude responses is TA = 64 ms. It can be observed
that the maximum energy transfer between the two modes occurs at the time
TA
4 = 16 ms. At this time, the x1 state achieves zero, x2 has a maximum ampli-
tude and vice versa. Accordingly, this fact suggests switch-off the parametric

10 -12
5
x
1
2.5 x
2
volts

-2.5

-5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)
16ms
10 -3
1
volts

0
Drive signal f
-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)
1
volts

0
Parametric excitation signal =0.72 volts
-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)

Fig. 6. Simulation for parameters in Table 1. Parametric excitation signal tuned at

ωp = ω2 − ω1 . The maximum transfer of energy occurs at the time T4A = 16 ms.
 Transient Vibrations Suppression in Parametrically Excited Resonators 203

excitation at T4A . Therefore, the time length of the pulse of parametric excita-
tion must be tp = T4A hence:
√
2π ω1 ω2
tp = (19)
βb

In this case, (γ = γ2 = γ1 ) the slow-motion frequency, does not depend on

γ, the damping term, γ is related to exponential behavior only. Thus, the length
of the pulse tp does not depend on γ either. From the formula (19) it can be
concluded that if the amplitude of parametric excitation increases, tp is reduced.
Therefore, to minimize the settling time in the first mode, the parameter βb must
be as large as possible. Thus, Fig. 7 and Fig. 8 show the time-response when the
system is coupled with a pulse of parametric excitation. Here, the settling time
of beam one was reduced from ts = 0.1s to ts = 16 ms, which means that the
settling time was minimized by 84%.

10 -12
5

x
2
2.5 x1
volts

-2.5

-5
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2

time (s) 16ms

10 -3
1
volts

0
Drive signal f
1
-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)

1
volts

Parametric excitation signal =0.72 volts

-1
0 0.02 0.04 0.06 0.08 0.1 0.12 0.14 0.16 0.18 0.2
time (s)

Fig. 7. Simulation for parameters in Table 1. Parametric excitation signal tuned at

ωp = ω2 − ω1 with the length of the pulse tp = 16 ms. The settling time of beam one
is minimized to ts = 16 ms approximately.
204 M. Ramı́rez-Barrios et al.

Fig. 8. Envelope of time-responses, for the parameters in Table 1. The top diagram
shows, that the period of the envelope amplitude is TA and the maximum transfer
of energy between the modes occurs at the time T4A = 16 ms. The inferior diagram
shows the amplitude envelope of the time responses when the system is excited by a
pulse of length tp = 16 ms. The settling time of beam one is minimized to ts = 16 ms
approximately.

6 Conclusions
Through the anti-resonance phenomenon description, a new method and pro-
cedure were presented to suppressing transient vibrations in High-Q microme-
chanical resonators. We obtain analytic formulas to tune properly a pulse of
parametric excitation signal to quickly transfer the response energy between the
modes of the system, herewith the responses vibrations in the first mode are
suppressed significantly, i.e., the settling time is minimized. The micromechan-
ical schemes allow for manufacturing an additional beam, i.e., an extra mode.
More research is needed to clarify if multiple frequencies of parametric excitation
signals are capable of transferring the energy into more modes and improving
the current results.

References
1. Blekhman I (2004) Selected Topics in Vibrational Mechanics, vol 11. World Scien-
tific, Singapore
2. Champneys A (2009) Dynamics of parametric excitation. In: Meyers R (ed) Ency-
clopedia of Complexity and Systems Science. Springer, New York, pp 2323–2345.
https://doi.org/10.1007/978-3-642-27737-5
 Transient Vibrations Suppression in Parametrically Excited Resonators 205

3. Collado J (2018) Hill equation: from 1 to 2 degrees of freedom. In: Clempner J, Yu

W (eds) New Perspectives and Applications of Modern Control Theory. Springer,
Cham, pp 43–71
4. Dohnal F (2008) General parametric stiffness excitation-anti-resonance frequency
and symmetry. Acta Mechanica 196(1–2):15–31
5. Dohnal F (2008) Damping by parametric stiffness excitation: resonance and anti-
resonance. J. Vibr. Control 14(5):669–688
6. Dohnal F (2009) Damping of Mechanical Vibrations by Parametric Excita-
tion: Parametric Resonance and Anti Excitation. Sudwestdeutscher Verlag fur
Hochschulschriften
7. Dohnal F, Ecker H, Springer H (2008) Enhanced damping of a cantilever beam by
axial parametric excitation. Archive Appl. Mechan. 78(12):935–947
8. Dohnal F, Verhulst F (2008) Averaging in vibration suppression by parametric
stiffness excitation. Nonlinear Dyn. 54(3):231–248
9. Nayfeh AH (2008) Perturbation Methods. Wiley, Hoboken
10. Okamoto H, Gourgout A, Chang CY, Onomitsu K, Mahboob I, Chang EY, Yam-
aguchi H (2013) Coherent phonon manipulation in coupled mechanical resonators.
Nat. Phys. 9(8):480
11. Okamoto H, Mahboob I, Onomitsu K, Yamaguchi H (2014) Rapid switching in
high-Q mechanical resonators. Appl. Phys. Lett. 105(8):083114
12. Ramı́rez M (2017) Attenuation of vibrations by parametric excitation (In Spanish).
Ph.D. thesis, Cinvestav
13. Ramı́rez M, Collado J (2016) Attenuation vibration by parametric excitation using
the meissner equation. In: 2016 13th International Conference on Electrical Engi-
neering, Computing Science and Automatic Control (CCE), pp. 1–6. IEEE
14. Ramı́rez M, Collado J, Dohnal F (2019) Coupled mathieu equations: γ-hamiltonian
and μ-symplectic. In: Awrejcewicz J (ed) Dynamical Systems Theory. IntechOpen
15. Ramı́rez Barrios M, Collado J, Dohnal F (2020) Stability of coupled and damped
mathieu equations utilizing symplectic properties. In: Lacarbonara W, Balachan-
dran B, Ma J, Tenreiro Machado J, Stepan G (eds) Nonlinear Dynamics of Struc-
tures, Systems and Devices. Springer, Cham. https://doi.org/10.1007/978-3-030-
34713-0 14
16. Thomsen JJ (2013) Vibrations and Stability: Advanced Theory, Analysis, and
Tools. Springer, Heidelberg
17. Tondl A (1998) To the problem of quenching self-excited vibrations. Acta technica
ČSAV 43(1):109–116
18. Verhulst F (2006) Nonlinear Differential Equations and Dynamical Systems.
Springer, Heidelberg. https://doi.org/10.1007/978-3-642-61453-8
19. Yakubovich V, Starzhinskii V (1975) Linear Differential Equations With Periodic
Coefficients, vol 1,2. Wiley, Hoboken
 Robust State Estimation for
Underactuated Systems Using Sliding
Modes and Attractive Ellipsoid Method

Bonifacio Sanchez1(B) , Patricio Ordaz2 , Ernesto Flores3 ,

and Alexander Poznyak1
1
CINVESTAV-IPN, 07360 Mexico City, Mexico
{bsanchez,apoznyak}@ctrl.cinvestav.mx
2
CITIS-UAEH, 42184 Pachuca Hidalgo, Mexico
jesus ordaz@uaeh.edu.mx
3
ESTi-UAEH, 43800 Tizayuca Hidalgo, Mexico
efloresg@uaeh.edu.mx

Abstract. The control of underactuated systems is a challenging and

interesting topic in applied engineering mainly due to their high-range
of applicability. This condition arises because they have less indepen-
dent control inputs than configuration variables. In real-time experimen-
tal environments, the presence of uncertainties, disturbances and the
unavailability of the full state vector complicate the control objectives.
Furthermore, when the number of passive joints is greater than the num-
ber of active ones, the stabilizability radius is too small, and the control
techniques based on the linear approximation are not robust enough to
perform stabilization in practice. Under these conditions, the control
problem when the full state-vector is not available becomes challeng-
ing. State-estimation may lead to a well-know peaking phenomenon. In
this paper, the Attractive Ellipsoid Method and Sliding Modes are used
to estimate the unavailable states. These give the possibility to reach
a stability-zone in finite time, while uncertain and the peaking phe-
nomenon effects are reduced, in order to provide sufficient conditions
for its applicability in observer-based control. Experimental results are
presented for the electromechanical triple-link inverted pendulum.

Keywords: Underactuated · Observer · Robust · Attractive Ellipsoid

Method · Sliding Mode

1 Introduction

The control design for underactuated systems is still an active field in engineer-
ing, mainly due to its high-range of applicability in complex systems. Walking
robots, mobile and aerial vehicles, elastic and flexible robots, pendular-type sys-
tems, Human-Robot interaction systems, mobile, aerial, water and underwater
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 206–217, 2020.
https://doi.org/10.1007/978-3-030-45402-9_20
 Robust State Estimation for Underactuated Systems Using Sliding Modes 207

robots, are common examples [1–8]. This class of systems has more configura-
tion variables than independent control inputs and put many challenging sit-
uations together. They are highly nonlinear systems and result to be control-
lable/stabilizable only in some regions of the state-space. For this reason, only
stabilization tasks are considered,such as practical stability [10,11]. This prob-
lem has been solved using many different techniques, but when the order of the
system increases, the problem may become too difficult in practice [9,16,17].
Furthermore, on a real-time experimental environment, model and parameters
uncertainties, matched and unmatched disturbances, the unavailability of some
of the internal variables, or the drawbacks of approximating them using tradi-
tional techniques such as filters, differentiators and dirty derivative [12,13] may
appear. In this sense, the estimator must respond faster than the natural fre-
quency of the plant, because, once the systems is initialized in some vicinity of
the considered solution, system may leave the attractive set before the control
acts, and the system may become unstabilizable. The problem turns to be more
difficult when uncertainties and disturbance effects are latent. This require of
robust estimation techniques, which leads to the peaking phenomenon [20]. High
undershoots/undershots on the estimated signals occur and lead to instability
when applied to an observer-based controller. A robust technique which allows
to deal with uncertainties/disturbances and this phenomenon at the same time
is thus required. The Attractive Ellipsoid Method is a robust technique which
concludes with Ultimately Uniformly Bounded stability [10,14]. As mentioned
in [10,11,15], this type of stability establishes that system’s trajectories reach an
invariant set in finite time, if they start from an interior point of a bounded set.
In this work, some robust Luemberguer-type observer structures are presented
for pendular-type underactuated mechanical systems driven by Permanent Mag-
net DC (PMDC) motors on a real-time experimental environment. The AEM
and Sliding Modes (SM) techniques are used for this aim. Results are presented
for the Triple-Link Inverted Pendulum (TLIP) system. It is important to remark
that this work has been partially submitted to the IEEE Transactions on Indus-
trial Electronics journal recently. See the reference [18] for details. Thus, the
main contribution of the work are some techniques to perform state-observation
for electromechanical underactuated systems driven by PMDC motors when
disturbances and uncertain effects are latent using the AEM and SM robust
techniques, while a reduction of the peaking phenomenon is achieved, in order
to provide the conditions for the observer-based stabilization in unstable con-
figurations under experimental environments. The following section presents the
problem statement in a formal manner, and some of the most important theoretic
preliminaries The state-space description of underactuated mechanical systems
driven by PMDC motors is presented in the quasi-linear format [9,14,18]. The
Sect. 3 presents the propositions referred to the observers design. Some remarks
about its applicability are also presented. Finally, Sect. 4 contain the experimen-
tal results for the electromechanical TLIP system, and last section describe the
concluding remarks.
208 B. Sanchez et al.

2 Preliminaries and Problem Statement

Consider an n−DoF Lagrangian system, coupled to the model of a set of m
independent Permanent Magnet DC (PMDC) motors driving its joints directly:

D(q)q̈ + C(q, q̇)q̇ + G(q) = WKτ Ia − τL ,

(1)
La I˙a + RaIa + Ke W q̇ = va , τL  ≤ δ0 ,
where q ∈ Rn is the generalised coordinates vector, Ia ∈ Rm the vector of
armature’s currents, va ∈ Rm the armature’s voltages, 0 < D(q) ∈ Rn×n the
inertia matrix, C(q, q̇) ∈ Rn×n the centrifugal and Coriolis forces matrix and
G(q) ∈ Rn the gravitational torques and forces vector. Furthermore, La ∈ Rm×m
is a diagonal matrix containing the armature’s inductances of the actuators,
Ra ∈ Rm×m the armature’s resistances Ke ∈ Rm×m and Kτ ∈ Rm×m the back-
emf and torque constants, respectively. The following properties are fulfilled for
the mechanical part, as exposed in [4,19]:
1. Inertia matrix is bounded in the following form:
λm In ≤ D(q) ≤ λM In , 0 < λm < λM , λm , λM ∈ R. (2)
p
2. Consider two vectors γ, χ ∈ R . The Coriolis and centripetal forces vector is
admitted to be radially unbounded with respect to χ as:
C(q, γ)χ ≤ δC γχ, 0 < δC , δC ∈ R. (3)
3. The gravitational torques and forces vector is bounded as
G(q) ≤ δG , 0 < δG , δG ∈ R. (4)
Let select the change of variables as x1 = Ia , x2 = q, x3 = q̇, such that x =

[x1 , x2 , x3 ] ∈ D ⊆ Rr , r = 2n + m is the state vector and ut = va ∈ U ⊆ Rm
the control input. The nonlinear control affine representation results in the form
ẋt = f (xt , t)+But +ϑ(xt , t) arises. Consider however the following remark about
the state-space quasi-linear representation.
Remark 1 (The quasi-linear representation). The linear approximation
around the equilibrium solution xeq ∈ D is found for the nominal
 system, i.e.,
∂f (xt ) 
when τL = 0, using Taylor’s series expansion as A = ∂xt  ∈ Rr×r . The
xt =xeq
so called quasi-linear format is obtained as
ẋt = Axt + But + ξ(xt , t), x(0) = x0 . (5)
For the particular case of electromechanical systems driven by PMDC motor:
⎡ ⎤ ⎡ ⎤  
−L−1 −1
a Ra 0m×n −La Ke W

L−1
a 0
A=⎣ 0n×m 0n×n In ⎦, B= 0n×m ⎦, ξ(xt ,t)=
⎣ 0 ,
−1
A31 A32 A33 0n×m o(xt )+D (x2 )τL (t)
(6)
o(·)=D −1
(·) {WKτ x1 − C(·)x3 − G(·)} − [A31 , A32 , A33 ] xt ,
where o(xt ) is given and contains the nonlinear terms of the nominal model, and
A31 ∈ Rn×m , A32 ∈ Rn×n and A33 ∈ Rn×m come from the linear approxima-
tion of the mechanical part.
 Robust State Estimation for Underactuated Systems Using Sliding Modes 209

For this representation, it results that:

 
Lemma 1 (On the boundedness of ξ(·)). Let define A3 = A31 , A33 , A32 .
The uncertainties, disturbances and nonlinear terms vector ξ(xt , t) ∈ Rr is quasi-
Lipschitz, i.e., it is bounded as
 √ √ 2
δ0 2 δ r
ξ(xt ,t)2 ≤γ0 +γ1 xt 2 , γ0 =2( λm ) , γ1 =2 λf m + λmax (A3 A3 ) (7)

for the scalar constant max{max(kτj ), δG , δc } ≤ δf .

Proof. Refer to [18].
Also consider the quasi-linear representation in the regular observable form:
Remark 2 (The regular observable form). Consider the non-singular coor-
dinate transformation
   
z1 Nc
zt = = Γxt = xt ,
z2 C

where Nc ∈ R(r−p)×r is the null space of the output matrix. The alternate rep-
¯ t , t), as
resentation is found to be żt = Āzt + B̄ut + ξ(z
        
ż1 Ā11 Ā12 z1 B̄1 ξ¯
= + u + ¯1 ,
ż2 Ā21 Ā22 z2 B̄2 t ξ2 (8)
y t = z2 ,

where Ā = ΓAΓ−1 , B̄ = ΓB and ξ(z ¯ t , t) = Γξ(xt , t). Henceforth, Ā11 ∈

R (r−p)×(r−p)
, Ā12 ∈ R (r−p)×p
, Ā21 ∈ Rp×(r−p) , Ā11 ∈ Rp×p , B̄1 ∈ R(r−p)×m ,
p×m ¯ r−p
B̄2 ∈ R , ξ1 ∈ R and ξ2 ∈ Rp . Here z1 ∈ Rr−p and z2 ∈ Rp contain the
¯
unavailable and measurable state-variables, respectively.
This lead to the following lemma.
Lemma 2 (On the boundedness of ξ(·)). ¯ The disturbances, uncertainties and
nonlinear terms vector ξ(xt , t) is quasi-Lipschitz, as seen in Lemma 2.1. This
condition may also be expressed in the new coordinates frame zt as:
 2
¯ λmax (Γ)
ξ(·) ≤ κ0 + κ1 zt  , κ0 = η0 λmax (Γ) , κ1 = η1
2 2 2
(9)
λmin (Γ)
Proof. Refer to [18].
The problem is now ready to be formulated.

Consider the case of a n−DoF mechanical system, driven by a set of

m−independent dc motors (1), in the quasi-linear format (5)–(6), or (8). The
problem is to estimate the unavailable state sub-vector z1 , using full or reduced
order observers, while disturbances and uncertain effects are latent, and provid-
ing a way for the reduction of the peaking phenomenon.
Consider the following theoretical preliminaries.
210 B. Sanchez et al.

Definition 1 (UUB stability [10]). The nonlinear dynamical system ẋ(t) =

f (t, x(t)), x(t0 ) = x0 , t ≥ t0 , is Ultimately Uniformly Bounded with bound b, if
there exists a > 0 such that, for every δ ∈ (0, a) there exists T = T (δ, b) > t0
such that :
x(t0 ) ≤ δ, =⇒ x(t) ≤ b, ∀t ≥ t0 + T. (10)
The following corollary [10], gives sufficient conditions to conclude with this.
Corollary 1. Consider the nonlinear dynamical system ẋ(t) = f (t, x(t)),
x(t0 ) = x0 , t ≥ t0 . Assume that there exist a proper function V : Rr → R
(i.e., continuously differentiable, positive definite and radially unbounded, [14]),
and class K functions (strictly increasing, [10]) α(·) and β(·) such that:

α(xt ) ≤ V (x) ≤ β(xt ), x ∈ D, V  (x)f (xt ) ≤ 0, x ∈ D, x ≥ μ (11)

where μ > 0 is such that Bα−1 (η) (0) ⊂ D for some η ≥ β(μ). Then the nonlinear
dynamical system is bounded. If last condition is now given as V  (x)f (xt ) <
0, x ∈ D, x > μ, where μ > 0 is such that Bα−1 (η) (0) ⊂ D for some η >
β(μ). Then the nonlinear dynamical system is ultimately bounded with bound
ε = α−1 (η). Furthermore, lim supt→∞ x(t) ≤ α−1 (β(μ)).
The AEM is closely related to this definition (see [14] for details). In order to
elucidate this method, consider that the nonlinear system is given in the quasi-
linear format (5). The AEM consists in designing a state-feedback controller
ut = Kxt , K ∈ Rm×r , such that UUB stability is concluded, for the Lyapunov
Candidate Function in the quadratic classic form V (xt ) = xt Pxt , 0 < P =
P ∈ Rr×r . In this sense, the ultimate attractive set has the particular form

E := {x ∈ Rr : x Patt xt ≤ 1} , Patt = μP.

where 0 < Patt ∈ Rr×r is the shape matrix of the ellipsoid. The length of the
semi-axes of the ellipsoid are backward proportional to the eigenvalues of Patt .
For this reason, the AEM is related to the solution of the following constrained
minimization problem.
minimize μ tr
1
P−1 by {P, K} ∈ Γ, subject to 0 < P = P , 0 < μ. (12)

where Γ is the set of restrictions that determine a class of admissible matrices

P, K.

3 Robust Observation
The following propositions are done to solve the stated problem. The first one
uses a full-order observer, with AEM.
Proposition 1. For the nonlinear system in the form (5)–(6), if there exist a
solution set (P, L|α, ε) for the matrix inequality:
PLC+C L P−PA−A P−αP−εη1 Ir , −P
Θ= −P , εIr > 0, (13)
 Robust State Estimation for Underactuated Systems Using Sliding Modes 211

with matrices 0 < P = P ∈ Rr×r , L ∈ Rr×p , and the scalars 0 < α, ε, then the
estimated state x̂t , from

x̂˙ t = Ax̂t + But + L(yt − ŷt ), x̂(0) = x̂0 ,

(14)
ŷt = C x̂t ,
converges to a invariant set about the real one xt , given by the ellipsoid:

E = {x̃t ∈ Rr : x̃t Patt x̃t ≤ 1} , Patt = (α/β)P,

and the so-called Ultimate Uniformly Bounded stability is concluded, with:

 
b= α β
λmin (P−1 ), T = α1 ln αV (x̃
ακ
0 )−β
,

for small enough 0 < κ << 1.

In the second one a reduced order observer was designed.
Proposition 2. Consider the electromechanical system in the quasi linear for-
mat (8). If there exists a solution set (α, ε|P, L) for the matrix inequality:
 PĀ +Ā P−PLĀ 
11 21
11
, PN
c −PLC
Θ= −Ā 
21 L P+αP+εη1 In−p <0 (15)
Nc P−C  L P , −εIn−p

then, the unavailable state sub-vector z1 ∈ Rr−p is approximated by the solution

of the Reduced Order Observer:
˙ t = Θ1 t + Θ2 yt + B̄1 − LB̄2 ut , (0) = 0 ,
ẑ1 = t + Lyt , Θ1 = Ā11 − LĀ21 , Θ2 = Ā11 L + Ā12 − LĀ21 L − LĀ22
without need of reconstructing the available one z2 ∈ Rp . Here, 0 < P ∈
R(r−p)×(r−p) and L ∈ R(r−p)×p are constant matrices, while α, ε ∈ R are positive
scalars. Then, the estimated state ẑ1 , converges to the ellipsoid:
 
E = z̃1 ∈ Rr−p : z̃1 Patt z̃1 ≤ 1 , Patt = (α/β)P.

Under this conditions, the so-called Ultimate Uniformly Bounded stability is con-
cluded, as in previous result.
The third approach joins AEM with Sliding Mode Observation.
Proposition 3. Let suggest the state observer [18]:

ẑ˙t = Āẑt + B̄ut + Lφ(zt ), ẑ(0) = ẑ0 ,

(16)
ŷt = C̄ ẑt ,
 
where L ∈ Rr×p is a constant matrix gain having the form L = L1  , Ip . If
the function φ(·) has the following structure:

φ(σt ) = Ā22 σt + ρ0 Sign(σt )+ ρ1 ,

(17)
0 < ρ0 ∈ Rp×p , ρ1 = η + δ4 λmax (Ā12 Ā12 ),
212 B. Sanchez et al.

then, in the region e1  ≤ δ4 , the estimated state variable ẑ2 converges to real
state variable z2 in finite time, given as:
√ 1
tr = trace(ρ0 )−1 2V12 (σ(t0 )) + t0 , (18)

where storage function V1 (σt ) is given in the quadratic form V1 (σt ) = 12 σt σt .
Now, consider the following storage function:

V2 (e1 ) = e1 Pe1 , 0 < P ∈ R(r−p)×(r−p) . (19)

Proposition 4. If for positive scalars α and β, there exists a solution of LMI:

⎛ ⎞
PA11 +A  
11 P−YA21 −A21 Y +αP , P , −Y
Θ=⎝ P , −εIp , 0 ⎠<0, P>0, Y=PL1 , (20)
−Y  , 0 , −εIp

is fulfilled, then for all t ≥ tr the storage function V2 (e1 ) satisfies:

β
V2 (e1 (t))≤ α +{V2 (e1 (tr ))− α
β
}e−α(t−tr ) (21)

where β = εη, and the Ultimate Uniform Bounded Stability is concluded, for:
√β 1

αV2 (tr )−β
 √
2
1
b= α λmax (P−1 ), T = α ln αζ + trace(ρ V 2 (σ0 )+t0 ,
0) 1
(22)

for small-enough constant ζ ∈ R+ .

Next section presents experimental results for the TLIP system.

4 Experimental Results
Experimental results were obtained for the Triple Link Inverted Pendulum sys-
tem of the PendCon Company, driven by a Maxon DC motor 218009 at first
joint. The instrumentation diagram is depicted in Fig. 1. The motor’s driver
4-Q-DC Servo-control LSC 30/2 and DAQ NI USB-6363 are used. System’s
numerical parameters are given in Table 1. Centre of masses were considered at
the centre of the link on each case. The dynamic model considers the mass of the
links and joints separately. With notation and parametric equations in Table 2,
inertia matrix is given as:
⎡ ⎤
D11 , D12 , m3 lc3 (l1 C31 +l2 C321 )
D(q)=⎣ D12 , Iω2 +(M2 +m3 )l22 , m3 l2 lc3 C321 ⎦
m3 lc3 (l1 C31 +l2 C321 ) , m3 l2 lc3 C321 , Iω 3

where:
D11 = Iω1 +Jm1 +(m1 +M1 +m2 +M2 +m3 )l12 +(M2 +m3 )l22 +2(M2 +m3 )l1 l2 C2 ,
D12 = m2 l1 lc2 C21 +l2 (M2 +m3 )(l2 +l1 C2 ).

Also, the centripetal-Coriolis forces matrix is found to be:

 
−2(M2 +m3 )l1 l2 S2 q̇2 , −θ1 q̇2 , −(θ2 +θ3 )q̇3
C(q,q̇)= θ1 q̇1 , 0 , −θ2 q̇3 ,
(θ2 +θ3 )q̇1 +θ2 q̇2 , θ2 (q̇1 +q̇2 ) , 0
 Robust State Estimation for Underactuated Systems Using Sliding Modes 213

Fig. 1. Experimental system block diagram.

Table 1. Numerical parameters of the Triple-Link Inverted Pendulum.

Parameter Value Parameter Value

Mass of the 1st link, m1 0.03 [kg] Length of the third link, l3 0.29 [m]
Mass of the 2nd link, m2 0.035 [kg] Armature’s resistance, Ra 2.44 [Ω]
Mass of the 3d link, m3 0.035 [kg] Armature’s inductance, La 612 [μH]
 
Mass of the 2nd joint, M1 0.1 [kg] Torque constant, Kτ 82.2 mNA
m
 
Mass of the 3d joint, M2 0.1 [kg] Back-emf constant, Ke 116 rpm
 V 2
Length of the 1st link, l1 0.19 [m] Inertia moment of the shaft, Jm 116 gcm
Length of the 2nd link, l2 0.26 [m]

and the gravitational torques and forces vector:

⎡ −m gl S −M gl S −m gθ −M gθ −m gθ ⎤
1 c1 1 1 1 1 2 4 2 5 3 6
G(q)=⎣ −m2 glc2 S12 −M2 gl2 S12 −m3 g (l2 S12 +lc3 S123 ) ⎦.
−m3 glc3 S123

From [19], centripetal-Coriolis forces matrix may be decomposed into the follow-
ing sub matrices:
     
0 −(M2 +m3 )l1 l2 S2 0 θ1 0 0 θ2 +θ3 θ2 0
C1 = −(M2 +m3 )l1 l2 S2 −θ1 0 ,C2 = 0 0 0 , C3 = θ2 θ2 0
0 0 −θ2 −θ3 , 0 0 −θ2 0 0 0

It follows that:

δc =32 max{(M2 +m3 )l1 l2 +m2 l1 lc2 ,m3 lc3 (l1 +l2 )} (23)

Table 2. Notation and parametric equations.

Notation Parametric equations

Sj = sin(qj ), Cj = cos(qj ) θ1 = (m2 l1 lc2 S21 + (M2 + m3 ) l1 l2 S2 )
Sjk = sin(qj + qk ), Cjk = cos(qj + qk ) θ2 = m3 l2 lc3 S321 , θ3 = m3 l1 lc3 S31
Skj = sin(qk − qj ), Ckj = sin(qk − qj ) θ4 = l1 S1 + lc2 S12 , θ5 = l1 S1 + l2 S12
θ6 = θ5 + lc3 S123
214 B. Sanchez et al.

3 1.2
Real data
Full-Order AEM 1
2
Reduced order observer

Armature Current Ia[A]

Joint AEM and SMO
0.8
1
0.6
0
0.4
-1
0.2

-2
0

-3
0 5 10 15 20 25 30 0 0.1 0.2
Time t[s]

Fig. 2. Estimation of the armature’s current.

On the other hand, with the Jacobian matrix of this vector respect to the gen-
eralized coordinate, it results evident that, with θ7 = m2 glc2 + M2 gl2 :

δG = 3θ7 max{m1 glc1 +(M1 +m2 +M2 )gl1 ,m3 g(l2 +lc3 )}, (24)

Results are presented under the knowledge of all the links positions. See
Figs. 2, 3, 4, 5, 6, 7 and 8. The gains are obtained using Matlab 2015 and cvx
2.1-toolbox, in Table 3.

Table 3. Numerical gains for the proposed observers.

α β L
⎡ −35.244 −0.443 −0.337 ⎤
12.64 −0.04 0.037
⎢ −0.04 12.72 −0.008 ⎥
Proposition 1 45 0.004 ⎢
⎣ 0.037 −0.008 12.84 ⎥ × 103
1109.5 22.06 4.55 ⎦
21.87 1147.5 −0.71
 2.00
4.6 −0.77 1158.5
0.034 0.0017
58.94 1.004 0.05
Proposition 2 45 0.004 1.00 60.14 0.001
 2.15 0.0034 60.125
0.05 0.001
0.0017
63.72 1.00 0.05
Proposition 3 65 0.004 1.02 65.07 0.0008
0.05 0.0008 65.06

The comparative survey is summarized in Table 4, where the parameters tr

and |M EV | are the rise-time (in milliseconds) and the magnitude of the mean-

0.5 0.2
Real data
0.4 Full-order AEM
Joint AEM & SMO
0.15
0.3

0.2
0.1
0.1

0
0.05
-0.1

-0.2 0

-0.3

-0.4 -0.05
0 5 10 15 20 25 30 0 2 4 6
-3
Time t[s] 10

Fig. 3. Estimation of the first pendulum’s position.

 Robust State Estimation for Underactuated Systems Using Sliding Modes 215

Table 4. Numerical results for the electromechanical TLIP system.

State-variable tr [ms] |M EV | × 10−3

Ia (t) 53.87 53.7 38.57 15 14.3 13.03
q1 (t) 0.173 — 2.85 0.008 — 0.007
q2 (t) 0.171 — 4.05 0.013 — 0.025
q3 (t) 0.169 — 2.84 0.008 — 0.008
q̇1 (t) 24 34.27 32.6 11 12.57 20.7
q̇2 (t) 25 42.34 39.8 1.33 4.12 11.64
q̇3 (t) 23 38.71 35.04 5.4 3.98 13.56
P1 P2 P3 P1 P2 P3

value (in the appropriate units) of the error’s signal, respectively. The tags P1,
P2 and P3 refer to the observers of propositions 1, 2 and 3 respectively.

0.3 0.3
Real data
0.25 Full-rder AEM
Joint AEM & SMO 0.25

0.2
0.2
0.15

0.1 0.15

0.05 0.1

0
0.05
-0.05
0
-0.1

-0.15 -0.05
0 5 10 15 20 25 30 0 2 4 6
Time t[s] 10-3

Fig. 4. Estimation of the second pendulum’s position.

0.3 0.3
Real data
0.25 Full-rder AEM
Joint AEM & SMO 0.25

0.2
0.2
0.15

0.1 0.15

0.05 0.1

0.05
-0.05
0
-0.1

-0.15 -0.05
0 5 10 15 20 25 30 0 2 4 6
-3
Time t[s] 10

Fig. 5. Estimation of the third pendulum’s position.

216 B. Sanchez et al.

6 2
Dirty derivative
Full order AEM
4 Reduced order observer 0
Joint AEM and SMO

2 -2

0 -4

-2 -6

-4 -8

-6 -10
0 2 4 6 8 10 12 14 16 18 20 0 0.5 1 1.5

Time t[s]

Fig. 6. Estimation of the first pendulum’s velocity.

3 2
Dirty derivative
Full-order AEM 0
2 Reduced order observer
Joint AEM and SMO
-2

1
-4

0 -6

-8
-1
-10

-2
-12

-3 -14
0 2 4 6 8 10 12 14 16 18 20 0 0.5 1 1.5
Time t[s]

Fig. 7. Estimation of the second pendulum’s velocity.

20 1
Dirty derivative
Full order AEM 0
15 Reduce order observer
Joint AEM and SMO -1
10
-2
5
-3

0 -4

-5
-5
-6
-10
-7
-15
-8

-20 -9
0 2 4 6 8 10 12 14 16 18 20 0 0.1 0.2
Time t[s]

Fig. 8. Estimation of the third pendulum’s velocity.

5 Concluding Remarks

Three robust techniques are presented to estimate the unavailable state vari-
ables for electromechanical systems. Experimental results were obtained for the
TLIP. When the observed signals are used to achieve state-feedback control
the problem becomes hard. For the experiments of this work, uncertainties were
introduced in system’s parameters and signals were estimated well. Finally it
is worth to mention that, observers tuning have been done with care of the
overshoot/undershoot effects in the transient response, in order to reduce the
dangerous peaking phenomenon. In future, this techniques will be tested on
real-time stabilization.
 Robust State Estimation for Underactuated Systems Using Sliding Modes 217

References
1. Olfati-Saber R (2002) Normal forms for underactuated mechanical systems with
symmetry. IEEE Trans. Autom. Control 47(2):305–308
2. Xin X, Liu Y (2014) Control Design and Analysis for Underactuated Robotic
Systems. Springer, Heidelberg
3. Fantoni I, Lozano R (2001) Non-linear Control for Underactuated Mechanical Sys-
tems. Springer, Heidelberg
4. Spong MW (1987) Modeling and control of elastic joint robots. J. Dyn. Syst. Meas.
Control 109(4):310–318
5. Ortega R, Spong MW, Gómez-Estern F, Blankenstein G (2002) Stabilization of a
class of underactuated mechanical systems via interconnection and damping assign-
ment. IEEE Trans. Autom. Control 47(8):1218–1233
6. Birglen L, Lalibertà T, Gosselin CM (2007) Underactuated Robotic Hands, vol
40. Springer, Heidelberg
7. Lefeber E, Pettersen KY, Nijmeijer H (2003) Tracking control of an underactuated
ship. IEEE Trans. Control Syst. Technol. 11(1):52–61
8. Reyhanoglu M, van der Schaft A, McClamroch NH, Kolmanovsky I (1999) Dynam-
ics and control of a class of underactuated mechanical systems. IEEE Trans.
Autom. Control 44(9):1663–1671
9. Bonifacio SR, Patricio OO, Alexander PG (2018) Robust stabilising control for
the electromechanical triple-link inverted pendulum system. IFAC-PapersOnLine
51(13):314–319
10. Haddad WM, Chellaboina V (2011) Nonlinear Dynamical Systems and Control: A
Lyapunov-based Approach. Princeton University Press, Princeton
11. Khalil HK (2014) Nonlinear Control. Pearson Higher Education, London
12. Ramirez-Neria M, Sira-Ramirez H, Garrido-Moctezuma R, Luviano-Juarez A
(2014) Linear active disturbance rejection control of underactuated systems: the
case of the Furuta pendulum. ISA Trans. 53(4):920–928
13. Alonge F, D’ippolito F, Raimondi FM, December 2001 Trajectory tracking of
underactuated underwater vehicles. In: Proceedings of the 40th IEEE Conference
on Decision and Control (Cat. No. 01CH37228), vol. 5, pp. 4421-4426. IEEE
14. Poznyak A, Polyakov A, Azhmyakov V (2014) Attractive Ellipsoids in Robust
Control. Birkhauser, Boston
15. Azhmyakov V, December 2011 On the geometric aspects of the invariant ellipsoid
method: application to the robust control design. In: 2011 50th IEEE Conference
on Decision and Control and European Control Conference, pp. 1353–1358. IEEE
16. Spong MW, September 1994 Partial feedback linearization of underactuated
mechanical systems. In: Proceedings of IEEE/RSJ International Conference on
Intelligent Robots and Systems (IROS 1994), vol. 1, pp. 314-321. IEEE
17. Lam S, Davison EJ, June 2006 The real stabilizability radius of the multi-link
inverted pendulum. In: 2006 American Control Conference, p. 6. IEEE
18. Sanchez B, Cuvas C, Ordaz P, Santos-Sanchez OJ, Poznyak A (2019) Full-Order
observer for a class of nonlinear systems with unmatched uncertainties: joint attrac-
tive ellipsoid and sliding mode concepts. IEEE Trans. Ind. Electron. 67:5677–5686
19. Kelly R, Davila VS, Perez JAL (2006) Control of Robot Manipulators in Joint
Space. Springer, Heidelberg
20. Khalil HK (2017) High-Gain Observers in Nonlinear Feedback Control, Society for
Industrial and Applied Mathematics (SIAM), Philadelphia
 Modelling and Control of 5 DOF Serial Robot
for Object Capturing and Manipulation Using
a Semi-analytical Approach

Bolaños Adrian(B) , Sánchez Kevin, Gabino Juan, Gutiérrez-Carmona Irandi,

and Martínez Manuel

Instituto Tecnológico de Monterrey Campus Tampico, Tamaulipas, Mexico

A01411983@itesm.mx, juan.diaz@tec.mx

Abstract. This paper details the kinematic analysis for a 5 degrees-of-freedom

serial robot, the Scorbot-ER III. Based on the Denavit-Hartenberg convention
is presented a semi-analytical relation between the end effector pose, and the
joints angles. The computational time required to calculate the robot trajectories
is reduced significantly, making this a suitable algorithm for real time applications;
specially for those where the robot is intended to move in a continuously changing
environment.

Keywords: Serial robot · Denavit-Hartenberg · Inverse kinematics · Forward

kinematics · Trajectories

1 Introduction
The capturing of space objects is an area with room for improvement; two applications
of this are the removal of space debris and on-orbit servicing (OOS). The former aims
to remove defunct artificially created free-flying objects that threats any manned or
unmanned spacecraft orbiting around Earth, while OOS missions are needed to repair
and give maintenance. In most cases the robot must interact with non-cooperative objects,
meaning that a constant analyzing of the motion of the environment is required in order
to calculate dynamical trajectories and identifying grappling points in the bodies [1].
The unpredictability and inconvenience of non-cooperative targets has led scientists
to propose different capturing methods using rigid or flexible connections [2]. One
flexible capture mechanism consists of a spacecraft equipped with an ejector that applies
acceleration to the masses on a net, afterwards the net expands and envelops the objective
allowing a secure capture [3]. Harpoons are also flexible capture mechanism that instead
of deploying a net, fires a harpoon that drills into the objective. Though both methods
have been studied and simulated, the dynamics on them are quite complicated and the
outcomes difficult to predict; furthermore, if the target is intended to be recovered without
damage, these previous methods cannot be implemented.
As an alternative, a rigid capture mechanism is considered [2]. Due to their versatility,
they are apt for capturing non-cooperative objects [4]. However, for the manipulator to
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 218–227, 2020.
https://doi.org/10.1007/978-3-030-45402-9_21
 Modelling and Control of 5 DOF Serial Robot 219

be considered a suitable option, a strategy to efficiently construct trajectories in real time

applications must be considered [5].
The problem of determining the actuators angle given the final position and ori-
entation of the end-effector is known as the inverse kinematic problem. The inverse
kinematic problem may be solved by numerical methods such as the Newton-Raphson
method. Unfortunately, the time required to find a feasible solution is not known due
to its dependency on arbitrarily set of initial conditions, and if a real-time trajectory
is to be determined using this method, the correct operation of the robot may be
compromised [6].
To circumvent this problem, artificial intelligence methods such as neural networks
are combined with other numerical methods to have more efficient and faster calcula-
tions [6]. Although artificial intelligence methods are the right tools when the model
is completely unknown or there are many uncertainties, one should resort to analytical
results if available.
In this work, we show that using the Denavit-Hartenberg (D-H) convention [7, p. 68],
a semi-analytical method can be easily found to solve the inverse kinematics problem
in a fixed time. The semi-analytical nature of our solution comes from the Scorbot-ER
III configuration, and the solution in a fixed time comes from the fact that a finite and
known number of iterations is needed to find a feasible solution.
This paper is organized as follows: In Sect. 2 is formulated the Scorbot-ER III
problem and defined the Denavit-Hartenberg matrices. In Sect. 3 the problem is solved
using a semi-analytical approach. In Sect. 4 is shown the numerical validation. And
finally, in Sect. 5 are shown the conclusions and future work is in Sect. 6.

2 Problem Formulation
The Scorbot-ER III robot is a versatile robot with 5 DOF and a working radius of
610 mm [8], designed to perform human tasks with basic actions that consist of gripper
open-close, body, shoulder and elbow rotations, and wrist roll and pitch movements [4].
Capturing space debris requires that, given a position in space, our robot calculates in
the smallest possible time a feasible trajectory to catch the flying object. Using numerical
methods such as Newton-Raphson methods, although effective and well-studied, does
not guarantee that the inverse kinematic problem will be solved fast enough. Hence,
the design of a simple algorithm that guarantees fixed convergence time for the inverse
kinematic problem is needed.

2.1 Denavit-Hartenberg Convention

The kinematic analysis of the Scorbot-ER III can be extremely complex, and the Denavit-
Hartenberg convention was introduced to provide a systematic procedure for performing
the analysis of the relationship between the individual joints of the robot manipulator
and the position and orientation of end-effector, simplifying the analysis considerably.
In Fig. 1 is shown the Scorbot-ER III initial configuration, and the corresponding
frames fixed to each of the articulations.
220 B. Adrian et al.

Fig. 1. Scorbot-ER III configuration and coordinate frames

⎡ ⎤
cθi −cαi sθi sαi sθi ai cθi
⎢ sθ cα cθ −sα cθ ai sθ ⎥
i−1
Ai = ⎢
⎣ 0
i i i i i i ⎥
(1)
sαi cαi di ⎦
0 0 0 1

In the Denavit-Hartenberg convention [7, 9], a homogeneous transformation matrix

i ∈ R
i−1 A 4×4 is used to describe the relation between the i-th and the (i − 1)-th

articulation,
where cξi = cos(ξi ), and sξi = sin(ξi ). Parameters αi , θi , ai , and di are described
in Table 1 [10].

Table 1. D-H parameters

System parameters
{X i , Yi , Z i } i-th coordinate frame
Qi Origin of the i-th coordinate frame
θi Joint angle, The rotation angle of each actuator
ai Link length, distance from Z i−1 to Z i with direction X i+
+
di Link offset, distance from Q i−1 to X i with direction Z i−1
αi Link twist, rotation from Z i−1 to Z i with respect to the X i -axis

The homogenous matrix i−1 Ai results from the multiplication of rotational and
translational matrices that are found analyzing the robot configuration and the actuators
positions, this is:
i−1
Ai = Rz,θi T ransz,di T ransx,ai Rx,αi (2)
⎡ ⎤
1 0 0 0
⎢ 0 cα −sα 0⎥
Rx,αi =⎢ i
⎣ 0 sα cα
i ⎥ (3)
i i 0⎦
0 0 0 1
 Modelling and Control of 5 DOF Serial Robot 221

⎡ ⎤
1 0 0 ai
⎢0 1 0 0 ⎥
T ransx,ai =⎢
⎣0 0 1 0 ⎦
⎥ (4)
000 1
⎡ ⎤
100 0
⎢0 1 0 0 ⎥
T ransz,di =⎢ ⎥
⎣ 0 0 1 di ⎦ (5)
000 1
⎡ ⎤
cθi −sθi 0 0
⎢ −sθ cθ 0 0 ⎥
Rz,θi =⎢
⎣ 0
i i ⎥ (6)
0 1 0⎦
0 0 01

The Denavit-Hartenberg parameters for the Scorbot-ER III are shown in Table 2.

Table 2. D-H parameters for the Scorbot-ER III

Coordinate Ai (cm) di (cm) Ai (degrees) θi (degrees)

frame
1 2.5 14 90 0
2 22 0 0 135
3 22 0 0 −135
4 0 0 90 0
5 0 14 0 0

2.2 Orientation and Position of End-Effector

The orientation of the end-effector can be described by defining the unit vectors of
{X 5 , Y5 , Z 5 } in terms of the inertial frame {X 0 Y0 Z 0 } by the transformation matrix,
0 T ∈ R4×4 (Fig. 2),
5
 0R
5 p5
0
T5 = (7)
01×3 1

where 0 R5 ∈ R3×3 , defines the orientation of the end-effector, this is

⎡ ⎤
cφ cγ cφ sα sγ − sφ cα sφ sα + cφ cα sγ
R50 = ⎣ sφ cγ cφ cα + sφ sα sγ sγ cα sγ − cφ sα ⎦ (8)
−sγ cγ sα cα cγ
222 B. Adrian et al.

Fig. 2. Unit vectors of end-effector in relation to the inertial frame

For simplicity, the elements of R50 may be rewritten as

⎡ ⎤
n x ox ax
R50 = ⎣ n y o y a y ⎦ = n o a (9)
n z oz az

where n, o, a ∈ R3 are defined by the direction cosines of the unit vectors {X 5 , Y5 , Z 5 },

in terms of the inertial frame {X 0 Y0 Z 0 }. Finally, the vector p5 ∈ R3 defines the position
of the end-effector, this is
⎡ ⎤
px
p5 = p y ⎦
⎣ (10)
pz

Matrix 0 T5 can also be defined as the transformation of the end-effector pose through
each coordinate frame using Eq. (1)
0
T5 = 0 A1 1 A2 2 A3 3 A4 4 A5 (11)

3 Main Results
By equating Eq. (9) with Eq. (11),
0
R 5 p5
= 0 A1 1 A2 2 A3 3 A4 4 A5 (12)
01×3 1

a set of nonlinear equations for the joints angles {θ1 , θ2 , θ3 , θ4 , θ5 } is found. In general,
due to the complexity of the equations, numerical approximations are required to deter-
mine the angles that satisfy Eq. (12). However, in the next sub-section, it is shown that
it may be found simple trigonometric relations to derive the θi angles.
 Modelling and Control of 5 DOF Serial Robot 223

3.1 Inverse Kinematics

In its construction, the Scorbot-ER III has 3 consecutive joints with parallel axes of
rotation. This results in the manipulator having one redundant angle. To avoid this
redundancy, θ2 was left as a free variable calculated by cycling through every physically
possible value and solving for each of the other angles using Eq. (12).
Pre-multiplying Eq. (12) by A−11 , the following trigonometric relations were found:

p y cθ1 = p y sθ1 (13)

a1 + a3 c{θ2 + θ3 } + a2 cθ2 + d5 ax cθ1 + a y sθ1 = px cθ1 + p y sθ1 (14)

d1 + a3 s{θ2 + θ3 } + a2 sθ2 = pz − d5 az (15)

sθ5 + n y cθ1 = n x sθ1 (16)

cθ5 + o y cθ1 = ox sθ1 (17)

Pre-multiplying by A−1 −1 −1
2 · A1 and post-multiplying by A5 the following simultaneous
equations were found:

s{θ3 + θ4 } = az sθ2 + ax cθ1 cθ2 + a y sθ1 cθ2 (18)

c{θ3 + θ4 } = sθ2 ax cθ1 + a y sθ1 − az cθ2 (19)

Using Eq. (13):

θ1 = atan2 p y , px (20)

From Eq. (14), (15):

θ3 = atan2( pz − d5 az − d1 − a2 s2 , px cθ1 + p y sθ1 − a1 − a2 cθ2

−d5 (ax cθ1 + a y sθ1 )) − θ2 (21)

Using Eq. (16), (17):

θ5 = atan2 n x sθ1 − n y cθ1 , ox sθ1 − o y cθ1 (22)

With Eq. (18), (19):

θ4 = atan2 cθ2 ax cθ1 + a y sθ1 + az sθ2 , sθ2 ax cθ1 + a y sθ1 − az cθ2 − θ3 (23)

Where atan2(y, x) returns the arc tangent of the two variable x and y, which is between
−π and π . Note that all angles are calculated using the inverse tangent, this is because
inverse sine and inverse cosine are prone to get indeterminations more often due to their
very limited domain and range.
224 B. Adrian et al.

Fig. 3. Points defined to create a circular trajectory

3.2 Trajectories
Trajectories were constructed by a set of points defining the position and orientation
of the final effector; this is, by construction the left-hand side of Eq. (12). Now, using
Eqs. (20), (21), (22) and (23) the set of the joints angles may be determined. Thus, the
Scorbot-ER III can move along a specific path maintaining the end effector on a variable
or fixed orientation (Fig. 3).

4 Numerical Validation
To prove the correctness of the inverse kinematics equations, the error norm
 0 
 R 5 p5 
 
 01×3 1 = − A1 A2 A3 A4 A5 
0 1 2 3 4
(24)
2

was observed for each point in the trajectory. The error norm was of the order of 10−2
units; this is, a highly accurate solution was found.
To evaluate the feasibility of the method based on its computational time, a circular
trajectory with center {40, 0, 20} cm and a 5 cm radius was computed (Fig. 4). For this
path, a total of 65 poses for the robot were calculated. The solution, shown in Table 3,
is found with a total computational time of 700 ms.

Table 3. Sets of angles for 5 points of the circle

θ1 (degrees) θ2 (degrees) θ3 (degrees) θ4 (degrees) θ5 (degrees)

7.125 70 −111.839 131.839 0
7.091 71 −111.532 130.532 0
6.989 72 −111.232 129.232 0
6.821 73 −110.948 127.948 0
6.588 74 −110.689 126.689 0
 Modelling and Control of 5 DOF Serial Robot 225

Fig. 4. Trajectory of a circle (top view)

It was found that the maximum variation between expected and calculated coordi-
nates was of 0.18 cm (Table 4).

Table 4. Comparison between approximated and real points

Approximated Real coordinates

coordinates (cm) (cm)
X Y Z X Y Z
40.1028 5.0128 19.9984 40 5 20
40.0746 4.9852 20.5042 40 4.9759 20.4901
40.0396 4.9084 21.0091 40 4.9039 20.9755
39.9956 4.7841 21.5099 40 4.7847 21.4514
39.9401 4.6128 22.0034 40 4.6194 21.9134

5 Conclusions

The Denavit-Hartenberg formalism allowed a simple analysis to determine the joints

angles in the Scorbot-ER III. Based on the transformation matrices and equations derived
from this convention, the time required to compute an approximated solution based on
a semi-analytical approach has proven to be considerably faster than numerical meth-
ods, making it a feasible solution to create trajectories with arbitrary paths. However,
physical limitations as the workspace and finite orientations of the end-effector must be
considered.
226 B. Adrian et al.

6 Further Projects
An autonomous robotic system can be created by implementing a set of camera and
distance sensors. This allows a continuous recognition of the target´s pose, creating
a 0 T5 matrix based in real on-time measurements. Then, the semi-analytical solution
proposed in this paper can be applied to find the solution in this changing environment.

Appendix
Scorbot-ER III transformation matrices.
⎡ ⎤
cos(θ1 ) 0 sin(θ1 ) a1 cos(θ1 )
⎢
0 A = ⎢ sin(θ1 ) 0 −cos(θ1 ) a1 sin(θ1 ) ⎥
⎥
1 ⎣ 0 1 0 d1 ⎦
0 0 0 1
⎡ ⎤
cos(θ2 ) −sin(θ2 ) 0 a2 cos(θ2 )
⎢
1 A = ⎢ sin(θ2 ) cos(θ2 ) 0 a2 sin(θ2 ) ⎥
⎥
2 ⎣ 0 0 1 0 ⎦
0 0 0 1
⎡ ⎤
cos(θ3 ) −sin(θ3 ) 0 a3 cos(θ3 )
⎢
2 A = ⎢ sin(θ3 ) cos(θ3 ) 0 a3 sin(θ3 ) ⎥
⎥
3 ⎣ 0 ⎦ (A1)
0 1 0
0 0 0 1
⎡ ⎤
cos(θ4 ) 0 sin(θ4 ) 0
⎢
3 A = ⎢ sin(θ4 ) 0 −cos(θ4 ) 0 ⎥
⎥
4 ⎣ 0 1 cos(α4 ) 0 ⎦
0 0 0 1
⎡ ⎤
cos(θ5 ) −sin(θ5 ) 0 0
⎢
4 A = ⎢ sin(θ5 ) cos(θ5 ) 0 0 ⎥
⎥
5 ⎣ 0 0 1 d5 ⎦
0 0 0 1

References
1. Xu W, Liang B, Li C, Xu Y (2010) Autonomous rendezvous and robotic capturing of non-
cooperative target. Robotica 28:705–718
2. Hausmann G, Wieser M, Haarmann R, Brito A, Meyer J-C., Jäkel S, Lavagna M, Jakobsson B,
Biesbroek R (2015) e. Deorbit Mission: OHB debris mission removal concepts. In: ASTRA
2015
3. Thomsen B, Sharf I (2016) Experiments on Tether-Net capture and net closing mechanism
of space debris. In: 67th International astronautical congress, Guadalajara, México
4. Ali MH, Aizat K, Yerkhan K, Zhandos T, Anuar O (2018) Vision-based robot manipulator
for industrial applications. Procedia Comput Sci 133:205–212
 Modelling and Control of 5 DOF Serial Robot 227

5. Basu R, Padage S (2017) Development of 5 DOF robot arm-gripper for sorting and
investigating RTM concepts. Mater Tod Proc 4(2):1634–1643
6. Rios A, Hernandez E, Moreno J, Keshtkar S, De la Garza R (2018) Kinematics analysis
of a new 3DOF parallel manipulator as walking rehabilitation device. In: 15th International
conference on electrical engineering, computing science and automatic control (CCE), pp
1–6
7. Zhu Q, Zhang Z (2017) An efficient numerical method for forward kinematics of parallel
robots. IEEE Access 99:1
8. Spong MW, Vidyasagar M (2008) Robot dynamics and control. Wiley, New York, p 68
9. Amatya S, Petchartee S (2015): Real time kinect based robotic arm manipulation with five
degree of freedom. In: The 2015 Asian conference on defence technology
10. Rosales EM, Gan Q (2002): Forward and inverse kinematics model for a 5-dof pioneer 2
robot arm. Essex
 Development of a Vision System for a Mobile
Manipulator to Solid Residue Collection
in the Food Sector

Jhoset Méndez Amador, P. A. Niño-Suárez(B) , and Raúl Rivera-Blas

Instituto Politécnico Nacional, ESIME Unidad Azcapotzalco, Ciudad de México, Mexico

joset_753@hotmail.com, {pninos,rriverab}@ipn.mx

Abstract. Due to its versatility, artificial vision has become very useful in differ-
ent areas, like medicine, where robotic arms and cameras have been incorporated
to carry out surgeries that involve a high level of precision. The combination
between the robotics and the artificial vision has managed to optimize the perfor-
mance of robotics arms in the object’s subjection. In this paper, the methodology
“V Model” is used for designing of the mobile manipulator, aided by algorithms
of artificial vision, to carry out the recognition of objects. These algorithms are
implemented on a FIT-PC4 embedded system. The object recognition algorithms
can calculate the position of the object in the Euclidean space and through arm
kinematics it can collect the objects and deposit them in a container inside the
mobile.

Keywords: V model · Artificial vision · Robotics · Inverse kinematics

1 Introduction

In food industry, equipment and facilities must be designed and built to ensure that the
safety of food can be guaranteed and prevent its contamination. It is essential that the
workspaces are always free of residues that can cause incidents in order to satisfy with
the highest quality parameters in the products. The food industry usually has a highly
competitive production, but with a relatively small robotic participation compared to the
manufacturing or automotive industries due to the products are highly variable in shape,
size and structure [1]. A viable alternative for incorporating the automation in cleaning
processes in industries (food sector), is the use of vision systems to support human on-site
supervision in the production cycle, in such a way that the information sent to the robotic
system be acquired without having any contact with the food that is being processed or
transported. Artificial vision can be defined as a field of artificial intelligence that, by
using the appropriate techniques, allows the collection, processing and analysis of any
type of special information obtained through digital images; being one of the fields of
research that has generated a lot of interest in recent decades. The inclusion of artificial
vision in the field of robot-assisted production has increased in recent years, with cameras
being the most suitable sensors for obtaining information about the robot’s environment,
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 228–239, 2020.
https://doi.org/10.1007/978-3-030-45402-9_22
 Development of a Vision System for a Mobile Manipulator 229

due to its wide field of panoramic view [2]. In the same way, they can also obtain
information about objectives, obstacles and even data that provide direct information
on the situation being displayed. The cameras offer the possibility of supporting mobile
robots in navigation tasks without the need for an operator [3–5].
The problem solved with the technological development presented, is common in
bakery companies; this occurs in production process, specifically when the bread goes
to the conveyor belt to continue the packaging process, there are times when the bread
falls to the ground and someone must be picking it up. There are multiple jobs where
artificial vision systems are used to support production processes; as presented by [6],
where an artificial vision system implemented in an embedded system for controlling
the operation of different types of wheeled mobile manipulators.
The solution devised supports medium and large bakery industries in cleaning tasks
at the production level, specifically, collecting solid waste on the soil and deposit them
in a predetermined place.
The vision algorithm identifies brown color object’s (hamburger bread), the object
is in the Cartesian coordinates of the mobile manipulator, and are sent to the platform
and robotic arm through the control algorithm; that way, the platform is moved to a
nearby point and then the arm is placed on the bread, so that it is recovered from the
soil and placed in the adequate container. Moreover, the incorporation of this kind of
technology in the production plants, will allow to carry out some of the routine and
heavy work previously destined for cleaning personnel, or as an additional task for the
plant operators, it will provide security in the production process because it does not
have human intervention in the food handling tasks.
In Sect. 2 the design of the vision system concept and operation control of the
mobile manipulator is presented, Sect. 3 presents the development of the vision and
control algorithms for operating the mobile platform, Sect. 4 presents some results of
laboratory experiments. Finally, there are the conclusions of the work carried out.

2 Design of the Vision System Concept

The system design was made by using the V-Model defined in [7], this tool presents a
general information flow for the product development process, since the idea conception
to its final design. It begins with the identification of user requirements, once these have
been taken into account a translation of the requirements is made into engineering terms
that will provide solution to needs; at the end, the V-Model ends with a system of user
validation, each stage of the product must be tested, in order to avoid incidents with
the customer, as mentioned in [8]. The initial part of the analysis, needs and customer
requirements, was carried out using the QFD Methodology, presented in [9], from these
results the design goals were obtained and from them a conceptual design was made, the
hardware architecture was choose and the control algorithms were developed by mean
of direct kinematic models of the manipulator and the mobile platform. These models
allow the mobile could move along a predefined path assigned by the user and stop
automatically when the camera detects the object of interest, then with the direct and
inverse kinematics of the robotic arm, the position to be reached is calculated.
230 J. M. Amador et al.

2.1 Design Goals

1. The vision system must be implemented in a mobile manipulator, according with

the following features:

a. The mobile platform must have wheeled locomotion.

b. The arm must be articulated and have a gripper.
c. The maximum load capacity of the mobile manipulator on its gripper, must be
in accordance with the task to perform.
d. The robot should move at a low speed (3 up to 5 m/s).
e. The platform dimensions must be: 500 mm wide, 600 mm long and 782 mm
high, or according to the workspace.
f. The minimum (arm retracted) and maximum (arm fully extended) robotic arm
reach and maximum must be in accordance with the task.

2. The vision and control system must provide the ability to acquire and process images
at the mobile manipulator, for making decisions during the trajectory and could be
autonomous in their movement.
3. Vision system must be able to operate in places with variable lighting.

2.2 Concept Generation

Based on the goals generated from the customer’s requirements, the function tree that the
vision and control system must possess for operating the mobile manipulator is defined
(see Fig. 1). The system detects the object to be collected (hamburger bread), indicates
to the mobile how far must move (to a point near the object), and the mobile manipulator
arm will perform the collection of the found object. The concept idea is to use the system
in general way to identify different types of objects and control different types of mobile
manipulators, by easily modifying some modules of the program.

Easy adaptation to new needs

Funcionality Mobile path control

Vision and control
system

Identifying objects by color

Energy Low power consumption

Fig. 1. Vision system function tree.

In Fig. 2a. the modules that make up the functional structure of the vision system
and operation control of the mobile manipulator are observed. The main module is the
digital processing system, in which the vision algorithms are implemented in order to
 Development of a Vision System for a Mobile Manipulator 231

give autonomy to the mobile manipulator. Similarly, a fundamental part of the vision
system is the selection of the camera for capturing the image to the digital processing
system. For the integral design of the vision system it is necessary to integrate the type
of actuators that generate the locomotion of the mobile manipulator and the robotic arm.

b. Mobile manipulator structure

a. Hardware architecture design proposal
proposal.

Fig. 2. Design proposal.

The captured images will be sent to the data acquisition system, where it will be
determined whether the robot should continue its trajectory or, if an object is detected,
evaluate whether it is the hamburger bread to be collected; therefore the distance to the
object is calculated to generate the control sequence that tells the actuators to move
the manipulator and collect it. As well, it is necessary to select the type of power for
supplying the necessary energy to the system, both for the mobile manipulator, as well
as for the vision system (camera and the data processing system).
As a result of the applied methodology and the customer’s requirements, a specific
structure of the mobile manipulator was selected for developing the mathematical mod-
els of the control module, being selected an anthropomorphic arm of three degrees of
freedom (DOF) and a mobile platform based on differential wheel. The proposed design
that involves the placement of the camera on the mobile platform and not on the manip-
ulator’s wrist (see Fig. 2. b), was evaluated together with the vision algorithm, because
the camera configuration does not move (see Fig. 2).
From the concept, the main modules for hardware and software architecture were
defined. It is important to remember that the system is modular, in this way the program-
ming of some modules could be modified to change the type of object to detected and/or
the mobile manipulator.

2.3 Hardware Architecture Design

The hardware architecture is a USB500W02M-AF30 camera and a FIT-PC4 embedded

system, with Linux operating system (64 bits) as work environment. The programming
software as the Eclipse IDE and the OpenCV vision libraries for image processing were
used, the main idea is to use free software tools.
232 J. M. Amador et al.

2.4 Mathematical Model

The robotic arm for the object manipulation is the Mitsubishi 5 DOF anthropomorphic
robotic arm RV-M1C, which has a clamp gripper that allows holding the object. The
Denavit-Hartenberg (DH) methodology to obtain its forward kinematic model was used
[10]. The reference systems of Mitsubishi manipulator could be observed in Fig. 3.

Fig. 3. Reference systems of Mitsubishi RV-M1C Robot.

The D-H parameters shown in Table 1, are substituted in Eq. (1) to create a homo-
geneous transformation matrix (HTM) for each DOF (Degree of Freedom) of the robot,
these individual matrices are multiplied to obtain the final HTM.

Table 1. Forward kinematic parameters.

N ai di αi θi
1 0 d1 90° θ1
2 a2 0 0 θ2
3 a3 0 0 θ3
4 0 0 90° θ4
5 0 d5 0 θ5

⎡ ⎤
cosθi −sinθi cosαi sinθi sinαi ai cosθi
⎢ sinθi cosθi cosαi −cosθi sinαi ai sinθi ⎥
⎢ ⎥ (1)
⎣ 0 sinαi cosαi di ⎦
0 0 0 1
Equation (2) corresponds to the H T M4×4 .
⎡ ⎤
a11 a12 a13 a14
⎢ a21 a22 a23 a24 ⎥
H T M4×4 = ⎢ ⎣ a31 a32
⎥ (2)
a33 a34 ⎦
a41 a42 a43 a44
where:
 Development of a Vision System for a Mobile Manipulator 233

a11 : −C1 S234 S5 − S1 C5 a22 : −S1 S234 C5 − C1 S5 a14 : C1 (l2 C2 + l3 C23 + l5 C234 )
a21 : −S1 S234 S5 + C1 C5 a32 : C234 C5 a24 : S1 (l2 C2 + l3 C23 + l5 C234 )
a31 : C234 S5 a13 : C1 C234 a34 : l1 + l2 S2 + l3 S23 + l5 S234
a41 , a42 , a43 : 0 a23 : S1 C234 a44 :1
a12 : −C1 S234 C5 + S1 S5 a33 : S234

To locate the position of the end effector on the Euclidean space, the position vector of
H T M4×4 is required (elements a14 , a24 , a34 ). This vector is sent to the control program
to perform the inverse kinematics of the manipulator, positioning the end effector where
the object is detected.
For modelling the movement of the mobile platform, the differential type wheeled
locomotion is used, its kinematic model is represented by the Eq. (3), where V is the
linear speed and W is the angular speed of the vehicle.
To obtain the angular velocity of each wheel (Wr , Wl ), the HTM of Eq. (4) is used,
by considering the values of wheel radius, r and the length between wheels spacing, l,
the wheel velocities are found.

ẋ = v ∗ cos(θ )
ẏ = v ∗ sin(θ )
θ̇ = w (3)
  
V r/2 r/2 Wr
= (4)
W r/2 ∗ l −r/2 ∗ l Wl

3 Vision Algorithm
The fundamental part of this work focuses on the vision algorithm, because it is useful
for the displacement of the mobile and object manipulation. The tests were carried out
by using different environmental conditions in order to avoid future problems for the
vision system.
The goal of the vision system is to locate an object; when it is located, the distance
between it and the camera is calculated, after it is transforms into Euclidian coordinates.
For adding robustness to the system, a brightness and contrast filter was implemented.
Figure 4 presents the general flow diagram of the system. The path tracking and
vision algorithms are executed at the same time, since they complement each other.
For example, if the vision algorithm is not working, the mobile manipulator will work
without receive control actions. On the other hand, if only the vision system is working,
the mobile manipulator will not move.
To achieve the bread detection, a method called segmentation was used, this is
based on the HSV (Hue Saturation Value) color model and a threshold technique. This
technique creates a binary image (black and white image), assigning a 1 in the image
where the object is detected and a 0 in the area of least interest. Color-based recognition
234 J. M. Amador et al.

Fig. 4. General flow chart of system operation.

is done by converting the original RGB (Red Green Blue) image to HSV, the Eq. (5)
performs a change of RGB values converting 8-bit values (0 to 255) to binary values (0
and 1). these values are transformed into HSV scale by the Eq. (6).
R  G B
R = G = , B =
255 255 255
Cmax = max R  , G  , B 
Cmin = max R  , G  , B 
 = Cmax − Cmin (5)
⎧    
⎪ ◦ × G −B mod6 , C 
⎪
⎪ 60  max = R
⎪
⎪  
⎪
⎪ ◦ B  −R 
+ 2 , Cmax = G  
⎨ 60 ×
    0, Cmax = 0
H= ◦ × R −G + 4 , C  S = 
⎪
⎪ 60  max = B Cmax , C max  = 0 (6)
⎪
⎪
⎪
⎪ 0◦
⎪
⎩
=0
V = Cmax
After the color conversion, the parameters range to detect the brown color of the bread are:
21 for Hue, 222 for Saturation and finally 202 for Value. In Fig. 5 the values established
for the brown color allow the proper identification of the object, highlighting its shape
in a green circle and another of red color indicating the position of the centroid.
To achieve the video binarization, a grayscale color conversion is performed; later,
the video is converted to black and white; finally, the binarized video is cleaned and any
possible noise is eliminated.
In this way, when non-zero values are detected, the edges of the object will be
displayed in white (see Fig. 6); this process is represented mathematically in Eq. (7).

max V al i f sr c > T (x, y)
dst(x, y) = (7)
i f not is equal to 0
In computer vision and image processing, the centroid (center of mass) is the weighted
average of the pixels that are the shape of an object. The location of the centroid in pixels
 Development of a Vision System for a Mobile Manipulator 235

is found by means of Eq. (8), using a relationship between the intensity of the pixels
on both X and Y axes over the total intensity of the pixels of the whole image. As the
position of the centroid is needed in cartesian coordinates, the Eq. (9) is used to perform
that task.
M10 M01
Cx = Cy = (8)
M00 M00
M10 M01
Posx = Pos y = (9)
Aobj Aobj

Using the instruction HoughCircles, based on the Hough Transform, the red circle indi-
cates the position of the centroid of the hamburger bread (see Fig. 7), and the green
circle highlights the shape of the object. In this way if the manipulator moves from one
place to another, both circles carry out the same action without losing the location of the
object.

Fig. 5. Correctly identified. Fig. 6. Binary image. Fig. 7. Hamburger bread

centroid.

The calculation of the distance is essential, since it is required to carry out the naviga-
tion and manipulation actions. The distance between camera and object was performed
by a technique called curve adjustment (Sum of Square Difference SSD), this method
is based on the area of the object and the camera used. The distance is calculated by
Eq. (10), where a multiplication is made between the size of the image in pixels, by the
inverse of the square root of the area of the identified object (−0.468). By increasing or
decreasing the area of the object the distance can change.
−0.468
distance = Pimg (Aimg ) (10)

Subsequently, the position of the object in Euclidean space must be calculated. First,
the intrinsic parameters of the camera and the reference systems to the base of the
manipulator and the camera are required. In order to obtain the intrinsic parameters
of the camera. The intrinsic parameters were obtained by the camera calibration, using
Matlab® and its Toolbox calibration. Table 2 shows the results of the camera calibration.
236 J. M. Amador et al.

Table 2. Intrinsic parameters of the USB500W02M-AF60 camera.

Focal distance f c = [345.688, 347.983]

Main point cc = [147.581, 98.604]

Before substituting values in Eq. (11), the rotation matrices and the translation of
the camera to the base of the robot are obtained by Eq. (12) and Eq. (13) respectively,
as mentioned in [11].
⎡ ⎤
⎡ ⎤ ⎡ −
→ −→ ⎤ X
λx p f k x r 1 + C x r 3 f k x t x + C x tz ⎢ ⎥
⎣ λy p ⎦ = ⎣ f k x −
→
r2 + C y −→
r3 f k y t y + C y tz ⎦⎢
Y ⎥
(11)
−
→ ⎣ Z⎦
λ r3 tz
1
⎡ ⎤
   0 −1 0
C R M = Rx(110◦ ) Rz(90◦ ) = ⎣ −0.3 0 −0.9 ⎦ (12)
+0.9 0 −0.3
⎡ ⎤ ⎡ ⎤
tx 0
cOM =⎣ ty ⎦ = ⎣ 0 ⎦ (13)
tz −tz

Using the complete model of Eq. (11), three equations are solved to obtain the points of
the object in Euclidean space. This X, Y, Z point is sent to the program that controls the
robotic arm, and the inverse kinematic is perform so that the bread be taken.
In Fig. 8 the initialization part of the vision algorithm is presented, where the
characteristics of the object to be detected are programmed.

Fig. 8. Vision system flow diagram, part 1.

 Development of a Vision System for a Mobile Manipulator 237

In Fig. 9 the algorithm continues with the application of the brightness and contrast
filter, the color conversion from RGB to HSV, the binarization of the video and the
calculation of the centroid of the object are performed. The final part of the vision
algorithm begins with the visualization of the object in the binarized image.

Fig. 9. Vision system flow diagram, part 2.

the distance between the camera and the identified object is calculated; finally, the
position of the object in X, Y, Z coordinates is obtained, and the arm control algorithm
is executed. The dotted line box is only executed when the user needs to make some
adjustment to the algorithm and wants to view changes on a monitor.

4 Results of the Application of the Vision and Control Algorithm

In Fig. 10 the result of one of the experiments is presented. In Fig. 10a, the initial position
of the mobile is displayed when the displacement trajectory starts to be executed, in this
case a straight line. The flexometer indicates the distance between the camera and the
bread. In the experiments corresponding to the first stage, the bread was placed at the
minimum (30 cm) and the maximum (70 cm) separation from the chamber. It was
observed that the technique used has an error of 1.6% (0.5 cm), with respect the real
measurement. This difference is acceptable for the type of application since it is not
necessary a high precision measurement. When the system is in spaces where the light
directly incidents on the object or in places with low lighting, the error is in the order of
2.3% (error 0.7 cm), which is still acceptable. From the results obtained, it was observed
that the segmentation technique has an accuracy of 98.3%, therefore it is a reliable
method.
In the second stage, where the kinematics of the mobile are involved, the manipulator
advances at 3 m/s once the bread is located (see Fig. 10b) and it stops at the minimum
separation distance (30 cm), which allows the manipulator to grab the bread. Although
there is a delay in milliseconds (10 ms) before the robot completely stops, this variation
does not affect the manipulation of the object (see Fig. 10c).
238 J. M. Amador et al.

a. Mobile b. Object detection and

c. Collection of the d. Placement of the
manipulator initial platform stop.
object. object in the container.
path.

Fig. 10. Result of one of the experiments

Finally, the execution time since the bread is taken until it’s and placed in the container
is 20 s, Fig. 10d. The workspace designated to execute the pick and place task was
appropriate and an arm with fewer degrees of freedom could be used, since this task is
easy to execute.

5 Conclusions
In this work, a vision system was developed for the recognition and location of an
object in Euclidean space. This was implemented in a system embedded in the mobile
robot. According with the defined problem to detect an object (hamburger bread) on the
production plant floor and the purpose of the task for the mobile manipulator was to
move to a point where it could be pick up and put it in a container. According to the
results obtained from the experiments realized, the precision of the system segmentation
technique is 98.3%.
The developed software architecture is modular, the user can modify the vision sys-
tem to recognize objects of different shape and size, and only the HSV values should be
adjusted depending on the color of the object. For this reason, the module offers versa-
tility and can be used in different tasks. In a similar way, the type of mobile manipulator
can be changed modifying the mathematical models of the robot (platform locomotion
type and the forward and inverse manipulator kinematics), in the corresponding module
of the software architecture.

Acknowledgements. We would like to thank to Instituto Politécnico Nacional for their support
through the Secretaría de Investigación y Posgrado. The first author thanks the Consejo Nacional
de Ciencia y Tecnología of México (CONACyT) for the scholarship awarded for Postgraduate
studies at the IPN-ESIME Azcapotzalco.

References
1. Chua, P.Y., Ilschnner, T., Caldwell, D.G.: Robotic manipulation of food products – a review.
Ind Robot Int J 30(4), 345–354 (2003)
 Development of a Vision System for a Mobile Manipulator 239

2. Arias M (2009) Navegación de un robot móvil en interiores utilizando flujo óptico a partir de
imágenes panorámicas, Tesis de Maestría, Universidad Nacional de Colombia, Medellín
3. Se S, Lowe D, Little J (2001) Vision-based mobile robot localization and mapping using scale-
invariant features. In: Proceedings of the 2001 IEEE international conference on robotics &
automation, Seoul, Korea, May 2001, pp 1–8
4. Graefe V (1989) Dynamic vision systems for autonomous mobile robots. In: IEEE/RSJ
international workshop on intelligent robots and systems 1989, Tsukuba, Japan, pp 1–12
5. Ishiguro H (1997) Distributed vision system: a perceptual information infrastructure for robot
navigation, Sakyo-ku, Kyoto, Japan
6. Bulanon, DM, Kataoka, T, Okamoto, H, Hata, S (2004) Development of a real-time machine
vision system for the apple harvesting robot. In: SICE annual conference in sapporo, Hokkaido
Institute of Technology, Japan, 4–6 August 2004, pp 1-4
7. Wang, Y, Ying, Y, Chun, X, Xiaoyang Z, Weizhi, J (2012) A proposed approach to mecha-
tronics design education: integrating design methodology, simulation with projects, Elsevier,
pp 942–948
8. Zheng, JL, Duigou, Bricogne, M, Eynard, B (2013) Survey of design process models for
mechatronic systems engineering. In: CIGI 2013, Compiègne Cedex, France, pp 1–6
9. Bergquist, K., Abeysekera, J.: Quality Function Deployment (QFD) - a means for developing
usable products. Int J Ind Ergon 18, 269–275 (1996)
10. Spong, MW., Vidyasagar, M (1989) Robot dynamics and control
11. Ibarra, JM, Iturbe, EJ (2017) Calibration and correction for radial distortion on vision sys-
tems for Robots, Robotic Laboratory, Control and Automation Department of CINVESTAV,
Mexico, DF
 Robust Control of a Master-Slave
Manipulator Under Restricted Task-Space

D. Cruz-Ortiz1(B) , I. Chairez2 , and A. Poznyak1

1
Automatic Control Department, CINVESTAV-IPN, Av. Instituto Politécnico
Nacional 2508, Gustavo A. Madero, 07360 Mexico City, Mexico
{dcruz,apoznyak}@ctrl.cinvestav.mx
2
Bioprocesses Department, UPIBI-IPN, Av. Acueducto 550, Gustavo A. Madero,
07340 Mexico City, Mexico
jchairezo@ipn.mx

Abstract. The aim of this manuscript is to develop a robust output-

based controller for a master-slave robot manipulator (RM), under state
constraints. The proposed controller satisfies a regular proportional-
derivative (PD) structure with time-varying gains. This controller
includes an integral compensation term which is guaranteeing the colli-
sion avoidance with possible obstacles placed within the boundaries of
the task work-space. The attractive ellipsoid method justifies the design
of the control strategy. The solution of a matrix inequality character-
izes the zone of convergence and the sub-optimal control gains. A set of
numerical simulations based on a two-link RM illustrates the advantages
obtained with the proposed method.

Keywords: Robust control · Constraint system · Attractive ellipsoid

1 Introduction

Robotic manipulators (RMs) are electrical-mechanical devices with diverse appli-

cations in researching and industrial sectors [2,12]. The RMs are usually pro-
grammed to perform a repetitive and autonomous sequence of movements. These
devices may exert tasks that can be dangerous by a human operator [1,5]. This
fact favors the interest on developing new manipulator configurations, as well
as, supplying them with new skills and degree of freedoms (dofs).
The most notable configuration of RM using the human experience and abil-
ity coupled with the potential of robotic systems [9] is a master-slave configura-
tion (MSC). Nowadays, the MSC has been used in several technical applications
due to its versatility for the movements of the final effector [6,13].

David Cruz-Ortiz is sponsored by Mexico scholarship from the CONACYT, scholar

reference: 550817.
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 240–248, 2020.
https://doi.org/10.1007/978-3-030-45402-9_23
Robust Control of a Master-Slave Manipulator Under Restricted Task-Space 241

The remote operation of RMs that realize their movements in a constrained

space has encouraged the developing of automatic controllers which can ensure
two conditions: (1) completing the tracking of a reference trajectory (provided
by a Master Robot (MR)) and (2) guaranteeing the avoidance collisions with
the objects located in the environment, as well as, satisfying the boundaries of
the workspace.
The trajectory tracking problem between the states of the MR and the slave
Robot (SR) can be solved by diverse controllers [7,10,11]. Linear feedback con-
trollers or sliding modes have been widely used to solve the tracking problem for
MSC [3,4]. However, these approaches may not satisfy the state constraints.
From the control point of view, notice that the RMs is a system affected
by bounded perturbations (external). Additionally, it is unrealistic to expect
that the mathematical models represent the nonlinear dynamics without uncer-
tainties. Considering the previous argument, there are several results to get the
minimization of the effects due to external perturbations and internal uncer-
tainties. One of the most remarkable result is the attractive ellipsoid method
(AEM) [8]. The AEM deals with unmatched perturbations. Also, the control
gain is usually obtained by the solution of a linear minimization problem with
a linear objective function and linear constraints.
This study develops a state feedback controller for MSC, aimed to satisfy
restrictions in the states in presence of model uncertainties and external pertur-
bations. The controller includes an integral compensation term to satisfy both
the state constrains and collision avoidance for the SR. The application of the
AEM estimates the gains of the controller.

1.1 Plant Description

Let define the dynamical model of the robotic manipulator, considering xi1 ∈ Rn
and xi2 ∈ Rn :
ẋi1 = xi,2

ẋi2 = Λi (xi ) + Θi (xi1 )ui + ηi (xi ) (1)

yi = Υi xi
  

where xi = x i1 xi2 , xi ∈ R2n . The subscript i = {m, s} describe the MR and
SR. The drift function Λi : R2n × R+ → Rn gathers all the terms associated with
gravitational effects and components of Coriolis matrix. This function admits the
following restriction

Λi (z1 ) − Λi (z2 )2 ≤ Li1 + Li2 z1 − z2 2 (2)

with z1 ∈ R2n and z2 ∈ R2n and Li1 and Li2 are positive constants.
The state dependent matrix Θi : Rn × R+ → Rn×n is a symmetric positive
matrix, that represents the robot inertial term [1]. The matrix Θi has an inverse
matrix Θi−1 ∀t ≥ 0, also has bounded norm Θi (z3 )2 ≤ gi+ .
242 D. Cruz-Ortiz et al.

with gi+ a positive scalar and z3 ∈ Rn . Because of the positiveness and symmetry
of matrix Θi , the fully actuated system concept applies for the master device.
Based on the previous fact, the following assumption is valid.

Assumption 1 The non-linear system (1) is controllable.

The term ηi : R2n × R+ → Rn represents the effect of external perturbations

and internal uncertainties and admits the upper bound

ηi (xi )2 ≤ Liη1 + Liη2 xi 2 (3)

where Liη1 and Liη2 which are positive constants.  

The vector yi ∈ Rn is the measurable output with Υi = In 0n ∈ Rn×2n
and the term ui ∈ Rn represents the control input. Here it is denoted the set
ΩM that represents the space where the trajectories of the MR are restricted.
The set Ω1 represents the state constraints for the SR. The obstacles may exist
within the workspace Ω1 . These obstacles correspond to Ω2 . Therefore, the real
workspace of SR is ΩW = Ω1 − Ω2 .

2 Problem Statement

Usually, the reference trajectories for the MR came from a human operator
(open-loop). However, in some cases, these reference trajectories can be obtained
as a result of a feasible trajectory tracking problem between the states of MR
and some desired reference trajectories (close-loop). This problem can be refor-
mulated as the design of the control input (um ) for the MR such that

um  {um : limxm1 − x∗m1  = 0 ∀t ≥ T } (4)

where x∗m1 ∈ Rn describes the reference trajectory of the MR. Here, it is consid-
ered that each component of x∗m1 is a C 1 function. The finite-time convergence
of this system is described by T .
From the physical point of view, the state xm1 of MR is bounded as a conse-
quence of mechanical restrictions in the RM. Therefore, it is possible to establish
the upper bounded from the state xm as a constant x+ m , such that:

xm ∈ dom{xm | xm 2 ≤ x+
m, m ∈R }
x+ +
(5)

Considering that the reference trajectories of SR are given by the movements

executed by the MR, the set of admissible trajectories for the SR must be upper
bounded by x+s , that is:

xs ∈ dom{xs | xs 2 ≤ x+
s , s ∈R }
x+ +
(6)

The SR trajectories must be restricted by the boundaries of ΩW , avoiding

the collision with the obstacles. The previous fact can be reformulated as a
Robust Control of a Master-Slave Manipulator Under Restricted Task-Space 243

trajectory tracking problem with state constraints defined as follows:

xm1 − xs1  ≤ β, β ∈ R+

subjected to (7)

xs1 − xf  > ,  ∈ R+
Here xf ∈ Rn defines the vector corresponding the distance between each
σ − ith component of state xs1 and the boundary ∂ΩW of the set ΩW . Notice
that ∂ΩW = ∂Ω1 ∪ ∂Ω2 where ∂Ω1 is the boundary of the set Ω1 and ∂Ω2
is the boundary of the set Ω2 . Also, 0 ∈ / ∂ΩW . Then, the norm used in the
inequality (7) corresponds to the Hausdorff distance, that is d(xs1 , ∂ΩW ) =
inf ζ∈∂ΩW {d(xs1 , ζ)}
The objective of the controller is designing the admissible control us ∈ U adm
aimed to solve the tracking trajectory between xm1 and xs1 fulfilling that
xs1 ⊂ ΩW , subject to the state constraint defined in (7). The control input
belongs to the admissible set U adm characterized as:
U adm = {us : us 2 ≤ us0 + us1 x2 }, us0 , us1 ∈ Rn (8)
The condition (8) allows to design an output feedback controller based on the
AEM [2].

3 Controller Design
3.1 Master Device
In this case of study the sliding mode controller is proposed to regulate the
trajectory tracking for the MR. However, this controller does not contemplate
the state constraints. The previous fact justifies the controller design with the
state constraints for the SR. The sliding mode controller for the MR is described
by the following expression
um = −Km Sign(s) (9)
The variable s ∈ Rn is the sliding surface defined by s = Δm + C̃ Δ̇m . The
variable Δm ∈ Rn , defined as Δm = xm1 − x∗m1 , describes the tracking error and
Δ̇m ∈ Rn is the derivative of the tracking error, with the matrices Km ∈ Rn×n
and C̃ ∈ Rn×n . The function Sign(s) satisfies the following definition:
 
Sign(s)  sign(s1 ) sign(s2 ) ... sign(sn ) (10)
The problem of the tracking trajectory between the master and the slave
devices can be solved by some well-developed control schemes. The most common
manner for solving the controller design is the so-called PD controller. However,
this kind of control, cannot solve the state constraints considered in the problem
formulation. In this section, a stable PD controller with state constrains can be
developed. This controller applied on the SR and allowed ensuring an adequate
trajectory tracking.
244 D. Cruz-Ortiz et al.

3.2 Slave Device

Let us consider the trajectory tracking error Δ ∈ R2n defined as Δ = xs − xm .

To satisfy simultaneously the control objective and the state constraint defined
in (7), suggest the control input us  us0 + us1 where us0 ∈ Rn is the nonlinear
compensation and us1 ∈ Rn is the term that ensures the state constraints. This
control is defined as us1 = Θs−1 ũs1 where ũs1 ∈ Rn satisfies:

ũs1 = −Φ1 Δ − Φ2 κ (11)

Here, the control gains Φ1 ∈ Rn×2n , Φ2 ∈ Rn×n should be designed. The term
κ ∈ Rn is proposed as κ = Λs + Θs us0 .

3.3 Stable Equilibrium Point for Tracking Trajectories

Consider the tracking error Δ, the full time derivative is Δ̇ = ẋs − ẋm , that can
be rewritten as Δ̇ = Λ(Δ) + Θs (us0 + us1 ) + η. The function Λ satisfy:
 
 χ
Λ(Δ) := χ Λ + AΔ + Δ. (12)
0n×2n

The matrix A ∈ R2n×2n is Hurwitz, the matrix χ ∈ Rn×2n has the following
structure χ = 0n In , the function Θs is Θs := χ Θs and the function η is
governed by: η := χ η − AΔ, with Λ := Λs − Λm − Θm um and η = ηs − ηm .
The control inputs with the control gain Φ0 ∈ Rn×2n are proposed, to design a
stabilizing control:
us1 = −Θs−1 (Φ1 Δ + Φ2 κ)
(13)
us0 = −Θs−1 (Λ + Φ0 Δ)
The next equation define the time varying gain matrices
ρ
Φ1 := [1 + sign(Δ P Θs Φ2 κ)]
2
(14)
2 
Φ2 := Δ ρ

where the gain ρ ∈ Rn×2n and P ∈ Rn×2n a positive definite matrix.

Considering the control gains and substituting the control input us , the ful-
time derivative of Δ satisfy

Δ̇ = ÃΔ − χ (Φ1 Δ + Φ2 κ) + η (15)

where the matrix à ∈ R2n×2n is also Hurwitz. This matrix is à = A +


χ −Φ

0 .
The next theorem, provides the sub-optimal values of the control gain guar-
anteeing a minimal invariant attractive ellipsoid.
Robust Control of a Master-Slave Manipulator Under Restricted Task-Space 245

Theorem 1. If there exist positive definite matrix P and ρ, as well as the pos-
itive constants α and  such that Γ(P, ρ | α, ) ≤ 0, then the set Es (P α
β ) is an
invariant attractive ellipsoid for the state trajectories, that is:
 
α
lim sup Δ P Δ≤1 (16)
t→∞ β

4 ) and the matrix Γ(P, ρ | α, ) is:

where β = (1 + 2v + + 2+
⎡ ⎤
Γ11 −0.5P χ ρ −P χ ρ P
⎢−0.5ρ χP −In 0 0 ⎥
⎢ ⎥ (17)
⎣ −ρ χP 0 −In 0 ⎦
P 0 0 −In

Γ11 = P (Ã + 0.5αI2n ) + (Ã + 0.5αI2n )P

(18)
−0.5(P χ ρ + ρ χP ) + I2n×2n (1 + 2v + + 2+
4)

4 Numerical Results
The numerical evaluation of the suggested controller considered the two-link
manipulator as benchmark system (Fig. 1). The dynamic model of this device
satisfies:
M (q)q̈ + C(q, q̇)q̇ + Pq (q) = u (19)

Fig. 1. Two-link manipulator.

 
The vector q ∈ Rn , defined as q = q1 q2 , characterizes the generalized
position of the mechanic system, the variables li and mi (i = {1, 2}) represent the
length and the mass of each link respectively. The term u represents the control
 
input with u = u1 u2 . The term g describes the gravitational acceleration.
246 D. Cruz-Ortiz et al.

The inertia matrix M (q) is:

 
α + α2 + 2α3 cos(q2 ) α2 + 2α3 cos(q2 )
M (q) = 1 (20)
α2 + 2α3 cos(q2 ) α2

whit α1 , α2 and α3 are defined as follows

α1 = (m1 + m2 )l12
α2 = m2 l22 (21)
α3 = m2 l1 l2

The matrix C(q, q̇) satisfies:

 
−q̇2 −q̇1 − q̇2
C(q, q̇) = α3 cos(q2 ) (22)
q̇1 0

And the matrix Pq (q) corresponds to

 
α4 cos(q1 ) + α5 cos(q1 + q2 )
Pq (q) = (23)
α5 cos(q1 + q2 )

where the variables α4 and α5 are:

α4 = (m1 + m2 )gl1 , α5 = m2 gl1 (24)

The complete master-slave robotic system uses two manipulators like the
one depicted in Fig. 1. Both manipulator systems are simulated considering the
following parameters m1 = 0.2 kg, m2 = 0.6 kg, l1 = 0.6 m, l2 = 0.3 m, g = 9.8 sm2 .
The application of the proposed controller for the SR case, considered the
matrix à as follows.
⎡ ⎤
0 0 0.95 0
⎢ 0 0 0 0.95 ⎥
à = ⎢
⎣−475 −38 −142.50 −38 ⎦
⎥ (25)
−38 −161.5 −36.10 −152

The output perturbation was suggested as follows:

 
ηs (t) = (0.005 + 0.15 sin(t)) 1 1 (26)

The initial conditions for the system are selected as

 
xs (0) = 0.1 0.5 (27)

Selecting the parameters α = 0.4, and  = 0.4, the matrix obtained from the
solution of (17) is
⎡ ⎤
136.4498 −3.7838 36.1858 −2.1982
⎢ −3.7838 172.1710 0.0170 11.5184 ⎥
P −1 = ⎢
⎣ 36.1858 0.0170 246.1828 9.3179 ⎦
⎥ (28)
−2.1982 11.5184 9.3179 228.2731
Robust Control of a Master-Slave Manipulator Under Restricted Task-Space 247

This value of the matrix P leads to the control gain:

 
−0.0026 −0.0002 −0.0001 −0.0002
ρ= (29)
−0.0000 −0.0013 −0.0001 −0.0010

Figure 2 depict the results obtained if the control proposed in this manuscript
is applied. The tracking performances comparison between the suggested con-
troller with state constraints (P DC ) and the classical PD. The performance and
the corresponding zoom for the tracking trajectory obtained for the first link of
SMR is depicted in Fig. 3, where qm,1 is the reference provided by the MR and
the constraints are defined by C1 and C2 .

Fig. 2. Tracking errors and attractive ellipsoid.

Fig. 3. Tracking trajectory of the first link of SR.

248 D. Cruz-Ortiz et al.

5 Conclusions
The controller design presented in this paper guarantees the trajectory track-
ing between two manipulator systems configured in a master-slave form. The
principal contribution of this paper is the inclusion of state constrains and col-
lision avoidance for the SR in the designed control based on the novel integral
compensation form.

References
1. Batlle J, Ridao P, Salvi J (1997) Integration of a teleoperated robotic arm with
vision systems using corba compatible software. In: 30th International Symposium
on Automotive Technology and Automation, Robotics, Motion and Machine Vision
in the Automotive Industries, pp 371–378
2. Capolei MC, Wu H, Andersen NA, Ravn O (2017) Positioning the laparoscopic
camera with industrial robot arm. In: 3rd International Conference on Control,
Automation and Robotics, pp 138–143. IEEE (2017)
3. Cruz-Ortiz D, Luviano-Juarez A, Chairez I (2014) Output sliding mode controller
to regulate the gait of gecko-inspired robot. In: XVI Congreso Latinoamericano de
Control Automático
4. Huang C, Xie L, Liu Y (2012) PD plus error-dependent integral nonlinear con-
trollers for robot manipulators with an uncertain jacobian matrix. ISA Trans
51(6):792–800
5. La HM, Lim RS, Basily BB, Gucunski N, Yi J, Maher A, Romero FA, Parvardeh
H (2013) Mechatronic systems design for an autonomous robotic system for high-
efficiency bridge deck inspection and evaluation. IEEE/ASME Trans Mechatron
18(6):1655–1664
6. Nguyen KD, Dankowicz H (2016) Cooperative control of networked robots on a
dynamic platform in the presence of communication delays. Int J Robust Nonlinear
Control 27:1433–1461
7. Pchelkin SS, Shiriaev AS, Robertsson A, Freidovich LB, Kolyubin SA, Paramonov
LV, Gusev SV (2017) On orbital stabilization for industrial manipulators: case
study in evaluating performances of modified PD+ and inverse dynamics con-
trollers. IEEE Trans Control Syst Technol 25(1):101–117
8. Poznyak A, Polyakov A, Azhmyakov V (2014) Attractive Ellipsoids in Robust
Control. Springer, Cham
9. Rebelo J, Schiele A (2012) Master-slave mapping and slave base placement opti-
mization for intuitive and kinematically robust direct teleoperation. In: 2012 12th
International Conference on Control, Automation and Systems (ICCAS), pp 2017–
2022. IEEE
10. Sharma R, Gaur P, Mittal A (2015) Performance analysis of two-degree of freedom
fractional order pid controllers for robotic manipulator with payload. ISA Trans
58:279–291
11. Su Y (2009) Global continuous finite-time tracking of robot manipulators. Int J
Robust Nonlinear Control 19(17):1871–1885
12. Tortora G, Dario P, Menciassi A (2014) Array of robots augmenting the kinematics
of endocavitary surgery. IEEE/ASME Trans Mechatron 19(6):1821–1829
13. Truong DQ, Truong BNM, Trung NT, Nahian SA, Ahn KK (2017) Force reflecting
joystick control for applications to bilateral teleoperation in construction machin-
ery. Int J Precis Eng Manuf 18(3):301–315
 On the Dynamics and Control
of a Single-Wheel Robot
with Inertial Locomotion

Mario E. Herrera-Cordero1 , Manuel Arias-Montiel1(B) , Marco Ceccarelli2 ,

and Esther Lugo-González1
1
Institute of Electronics and Mechatronics, Universidad Tecnológica de la Mixteca,
Huajuapan de León, Oaxaca 69000, México
herrera.enrique09@gmail.com, {mam,elugog}@mixteco.utm.mx
2
LARM2: Laboratory of Robot Mechatronics,
University of Rome, Tor Vergata 00133, Italy
marco.ceccarelli@uniroma2.it

Abstract. This work presents the mathematical modeling and the con-
trol design of a single-wheel robot with inertial locomotion. The forward-
backward robot motion is obtained by the controlled angular displace-
ment of a single pendulum, while the robot’s mathematical model is
obtained by using the Euler-Lagrange formalism. The resulting mathe-
matical model is represented by highly coupled non-linear differential
equations which are numerically solved and the obtained results are
validated by comparison with data from SolidWorks simulations. The
mathematical model is linearized around the operation point in order
to synthesize a control law to regulate the linear velocity of the robot.
The control scheme is based on the root locus method for the linearized
model. The control system performance is evaluated by numerical simu-
lations of the non-linear model in closed-loop. The obtained results show
the viability for the experimental implementation of the proposed control
system which is proposed as a future work for this research.

Keywords: Single-wheel robot · Dynamic model · Velocity control

1 Introduction
Single-wheel and unicycle robots have emerged as an alternative to multi-wheel
robots in order to reduce the size and the required space for operating. Single-
wheel robots are compact and they are able to operate in a flexible manner as
there is only one point of contact with the surface. This kind of robot can achieve
zero rotation radius and has potential applications, such as free movement and
load transportation in narrow spaces [1]. One of the most important issues in
single-wheel robots is the problem of lateral equilibrium. Different approaches
have been proposed to solve this problem: horizontal inertial wheels [2], lateral
c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 249–260, 2020.
https://doi.org/10.1007/978-3-030-45402-9_24
250 M. E. Herrera-Cordero et al.

inertial wheels [3] and high-speed gyroscopes [4]. Control strategies for single-
wheel and unicycle robots are first developed for straight trajectories as estab-
lished by Lee et al. [5]. In the context of single-wheel robots, a representative
investigation about the design and control of single-wheel robots is the work of
Jung et al. In [6] authors proposed a practical mechatronics approach for the
development of a the GYROBO. This mechatronic approach solves some prob-
lems regarding the system integration and allows the implementation of linear
controllers for both roll and yaw angles. In [7], Lee and Jung analysed the stabil-
ity of the vertical mode of GYROBO and test the balancing control performance
of GYROBO empirically to verify the comparison between the tested stability
bandwidth and the determined offset range. They only apply control for the
balancing or the GYROBO.
This paper is an extension of the work presented in [8] and presents the math-
ematical modeling and the control design of a single-wheel robot with inertial
locomotion. The system dynamics is analyzed by an energetic approach and the
robot’s mathematical model is obtained by using the Euler-Lagrange formalism.
The resulting mathematical model is represented by highly coupled non-linear
differential equations which are numerically solved and the obtained results are
validated by comparison with data from SolidWorks simulations. The mathe-
matical model is linearized around the operation point in order to synthesize a
control law to regulate the linear velocity of the robot. The control scheme is
based on the root locus method for the linearized model. The control system
performance is evaluated by numerical simulations of the non-linear model in
closed-loop. The obtained results show the viability for the experimental imple-
mentation of the proposed control system which is proposed as a future work for
this research.

2 Robot Description
The proposed design is based on the barycenter offset (BCO) driving principle.
In this kind of system, the mass distribution of the robots is shifted by an internal
mechanism in order to move the robot from one equilibrium position to another.
This principle has been implemented in diverse locomotion systems for spherical
robots [9].
In this project phase, the robot is only designed for forward and backward
movement on a straight trajectory and the mass distribution is shifted by the
movement of an inertial pendulum driven by a CD micromotor and internal spur
gears mechanisms. The mechanical elements of the single-wheel robot are shown
in Fig. 1. A more detailed description of the robot design and operation principle
is given in [8].

3 System Dynamics
The dynamic model of the single-wheel robot is obtained based on mechanical
elements shown in Fig. 1 and the following assumptions:
 On the Dynamics and Control of a Single-Wheel Robot 251

Eigth bolts separated

symmetrically about
Case
the axis.
(A disc and a tube)

R_c

Two base discs.

L_q

Axis

L_t
cg
There is no slip
between the
Little friction
wheel and the
between the disc
ground.
elements and base
pendulum. Sphere and stem are
considered a piece.

Fig. 1. Single-wheel pendulum robot configuration.

1. No slipping between the wheel and ground.

2. The pendulum is formed by a sphere and a solid rod, as they are considered
as one solid body.
3. Eight bolts distributed symmetrically with respect to the rotation axis are
considered.
4. Friction between the two base discs and pendulum is neglected.
5. The mass and moment of inertia of the pinion is neglected as these values are
considerably smaller than the values for the other mechanical elements.

In Fig. 2 a free-body diagram of a simplified view of the system is shown,

where θ is the robot angular displacement, τ is the input torque provided by the

Fig. 2. Free-body diagram of the single-wheel pendulum robot.

252 M. E. Herrera-Cordero et al.

micromotor, θ2 = θ − Rα is the pendulum angle, α is the pinion angular displace-

ment and R is the transmission ratio of the internal spur gears mechanism.
The linear velocity of the robot is

Ẋ = Rc θ̇ (1)

where Rc is the housing radius and θ̇ is the robot angular velocity.

To obtain the system’s kinetic and potential energies, we must analyse the
elements depending on the variable θ and separately, the elements depending on
θ and α as follows.

3.1 Elements Depending on θ

The elements depending on the variable θ are the robot housing, the base discs,
the bolts and the internal gears. Their kinetic and potential energies are given
by:
1
T1 = [mc Rc2 + md Rc2 + mcor Rc2 + Ic + Icor + 8(mbolt (L2q + Rc2 ) + Ibolt )]θ̇2 (2)
2

V1 = mc gRc + md g(Rc − Rd ) + mcor g(Rc − Rcor ) + 8mbolt gLq θ̇2 (3)

where Lq is the distance from the bolt axis to the rotation axis of the wheel,
mc is the robot housing mass, md is the base discs mass, mcor is the internal
gears mass, mbolt is a bolt mass and Ic , Id , Icor and Ibolt denote the respective
moment of inertia.

3.2 Elements Depending on α and θ

The elements depending on the variables α and θ are the pendulum and its base
(the composed pendulum) with mass center in Ctm given by the coordinates
Xctm and Yctm (see Fig. 3)
The kinetic and potential energies of the pendulum are [8]
1 1 2 1 2
T2 = mcomp Rc2 θ̇2 + mcomp ẊDctm cos(θ2 )θ˙2 + mcomp Dctm
2
θ˙2 + Ictm θ˙2 (4)
2 2 2
V2 = mcomp gDctm (1 − cos(θ2 )) (5)
where mcomp is the mass of the composed pendulum and Ictm is the moment of
inertia of the pendulum.
 On the Dynamics and Control of a Single-Wheel Robot 253

Fig. 3. Variables for the calculation of kinetic and potential energies of the pendulum.

3.3 Dynamic Model

The total kinetic and potential energies of the single-wheel robot are obtained
by Eqs. (2), (4), (3) and (5), respectively.
1
Ttotal = [mc Rc2 + md Rc2 + mcor Rc2 + Ic + Icor + 8(mbolt (L2q + Rc2 ) + Ibolt )
2
1 2 1 2
+ mcomp Rc2 ]θ̇2 + mcomp ẊDctm cos(θ2 )θ˙2 + mcomp Dctm2
θ˙2 + Ictm θ˙2
2 2
(6)
V1 = mc gRc + md g(Rc − Rd ) + mcor g(Rc − Rcor ) + 8mbolt gLq θ̇2
(7)
+ mcomp gDctm (1 − cos(θ2 ))
The Euler-Lagrange procedure can be applied to obtain the dynamic model
of the single-wheel pendulum robot described by the following expressions:
θ̈ = [mcomp Dctm Rc sin(bα − θ)(bα̇ + θ̇)2 + mcomp gDcomp sin(bα − θ)
1 (8)
− [mcomp Dctm Rc cos(bα − θ) − Icomp − mcomp Dcomp2
]bα̈ − cθ̇]
W
Tm
α̈ = [ − [mcomp Dctm Rc cos(bα − θ) − Ip − mcomp Dctm2
]θ̈
b
1 (9)
− mcomp gDctm sin(bα − θ)]
[Icomp + mcomp Dctm
2 ]b

where Tm is the input torque provided by the motor, c is the friction coefficient
Rpinion
between the wheel and the ground, b = R cor
, W = e−2mcomp Dctm Rc cos(bα−
θ) + Icomp + mcomp Dctm and e = mc Rc + md Rc2 + mcor Rc2 + 4(mbolt (L2q + Rc2 )) +
2 2

Ic + Id + Icor + 8Ibolt .
254 M. E. Herrera-Cordero et al.

3.4 Model Validation

For the model validation, Eqs. (8) and (9) were numerically solved with the
physical properties shown in Table 1. The obtained results were compared with
a simulation carried out in Solidworks software where the material properties
and friction parameters were assigned for a realistic simulation. In Fig. 4 the
obtained results for the linear velocity of the single-wheel pendulum robot with
a constant input torque of 2N · mm are shown.

Table 1. The physical properties of the single-wheel pendulum robot’s mechanical

elements.

Element Mass (g) Moment of inertia (g · cm2 ) Radius (cm)

Housing 583.44 73884.32 13.5
Base discs 70.42 1548.73 –
Bolts 1.87 0.35 1.64
Pendulum 122.36 586.611 10.3441
Pinion Negligible Negligible 0.75
Gear 4.199 7112.644 3.975

Fig. 4. Robot linear velocity with a constant input torque, numerical solution of the
non-linear equations and SolidWorks simulation.

As can be observed in Fig. 4, the results obtained by evaluating Eqs. (8) and
(9) are consistent with the numerical simulation carried out in the Solidworks
environment which is interpreted as the dynamic model validation.
 On the Dynamics and Control of a Single-Wheel Robot 255

4 Linearization, Transfer Function and Velocity Control

From Eqs. (8) and (9), it is possible to linearize the system model around an
operating point. This point is considered with the pendulum in a vertical position
down (which is also an equilibrium position), this implies θ2 = 0 and θ = bα.
Moreover, the equilibrium input is 0 and as a consequence the system has a
constant velocity, θ̈ = 0 and α̈ = 0. The state variables are defined as x1 = θeq ,
x2 = θeq˙ , x3 = αeq , x2 = α˙eq and Ueq = 0.
With the equilibrium conditions, it is possible to linearize the model by using
the Taylor series expansion [10] and considering the numerical values of Table 1,
the linearized model in space state is given by
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
x˙1 δ 0 1 0 0 x1δ 0
⎢x˙2δ ⎥ ⎢ −3.7951 −0.3034 0.7161 0⎥ ⎢x2δ ⎥ ⎢ −32.3079 ⎥
⎢ ⎥=⎢ ⎥⎢ ⎥ + ⎢ ⎥[Tm ]
⎣x˙3δ ⎦ ⎣ 0 0 0 1⎦ ⎣x3δ ⎦ ⎣ 0 ⎦ (10)
x˙4 δ 486.0901 0.4006 −91.7151 0 x4δ 2.0577e04
 
M atrixA M atrixB
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
y1 1000 x1δ 0
⎢y2 ⎥ ⎢0 1 0 0⎥ ⎢x2δ ⎥ ⎢ ⎥
⎢ ⎥=⎢ ⎥ ⎢ ⎥ + ⎢0⎥ [Tm ] (11)
⎣y3 ⎦ ⎣0 0 1 0⎦ ⎣x3δ ⎦ ⎣0⎦
y4 0001 x4δ 0
 
M atrixC M atrixD

From the linearized system of Eqs. (10) and (11) the transfer function can
be obtained as
G(s) = C(sI − A)−1 B + D (12)
⎡ ⎤
−32.3079s2 +1.1772e+04
s4 +0.3034s3 +95.5102s2 +27.5410s
⎢ ⎥
⎢ ⎥
⎢ 2
−32.3079s +1.1772e+04 ⎥
⎢ s3 +0.3034s2 +95.5102s+27.5410 ⎥
⎢ ⎥
G(s) = ⎢ ⎥ (13)
⎢ 2.0577e+04s2 +6.2302e+03s62387 ⎥
⎢ s4 +0.3034s3 +95.5102s2 +27.5410s ⎥
⎢ ⎥
⎣ ⎦
2.0577e+04s2 +6.2302e+03s62387
s3 +0.3034s2 +95.5102s+27.5410
It can noted that the second row from the right side of Eq. (13) corresponds
to the transfer function of single-wheel robot velocity.

−32.3079s2 + 1.1772e + 04
G(sθ̇ ) = (14)
s3 + 0.3034s2 + 95.5102s + 27.5410
The transfer function given by (14) will be used to develop the control scheme
for the single-wheel robot velocity. In order to validate this function, a numerical
simulation with a constant input torque of 2N · mm is carried out. The obtained
256 M. E. Herrera-Cordero et al.

Fig. 5. Comparison of linear velocity of the single-wheel pendulum robot with a con-
stant input torque, numerical solution of the non-linear equations and transfer function
response.

results are compared with the non-linear system response and this comparison
is presented in Fig. 5.
As can be appreciated, the obtained response from the transfer function
is practically the same as that the obtained from non-linear equations. These
results validate the linearization process and justify the use of the transfer func-
tion to synthesize the velocity control by linear techniques.
Before designing the controller, it is necessary to analyse the pole and zeros
placement in open and closed loop. In Fig. 6 the distribution of pole and zeros in
open loop is depicted. The open loop system is stable due to both poles having
negative real parts.

20
Root Locus of system in open loop
15 Pole: -0.00752 + 9.77i
Imaginary Axis (seconds -1)

Gain: 0
10
Zero: -19.1 Pole: -0.288
Zero: 19.1
5 Gain: Inf Gain: 0
Gain: Inf

-5 Pole: -0.00752 - 9.77i

Gain: 0

-10

-15

-20
-40 -20 0 20 40 60 80
Real Axis (seconds -1)

Fig. 6. Pole and zeros placement in open loop.

 On the Dynamics and Control of a Single-Wheel Robot 257

The closed loop transfer function is obtained as

C(s) G(s) −32.31s2 + 11772

= = 3 (15)
R(s) 1 + G(s)H(s) s − 32s2 + 95.51s + 11800

From the transfer function (15) it is possible to obtain the placement of pole
and zeros in closed loop which is shown in Fig. 7. There is a pair of conjugated
poles in the right semi-plane, which means the closed loop system is unstable
even if there are variations in the gain.

20
Pole: 23.4 + 15.6i Root Locus of system in closed loop
Gain: 0
15
Imaginary Axis (seconds -1)

10

5
Zero: -19.1 Zero: 19.1
Gain: Inf Gain: Inf
0
Pole: -14.9
-5 Gain: 0

-10
Pole: 23.4 - 15.6i
-15 Gain: 0

-20
-40 -20 0 20 40 60 80
Real Axis (seconds -1)

Fig. 7. Pole and zeros placement in closed loop.

It is possible to apply the root-locus control technique in order to impose

a desired behavior in the closed loop system. In this case, the proposed values
for the damping coefficient and natural frequency are ζ = 0.5 and ωn = 1 rad s ,
respectively. With these values, the dominant poles are defined by s1,2 = ζωn ±
1 − ζ 2 i ⇒ s1,2 = −0.5 ± 0.866i.
The closed loop instability is caused by the dominant pole in the right semi-
plane and the zero in the left semi-plane (see Fig. 7). A direct compensator is
proposed to cancel these pair of conjugated poles and the zero in the left side of
the complex plane.

−32.3079s2 + 11772
G(s) ∗ G(s)supr =
s3 + 0.3034s2 + 95.5102s + 27.5410
s2 + 0.01503027s + 95.49204
∗
(s + 19.088454)(s)
−32.3079(s − 19.088454)
= (16)
(s + 0.288)(s)
The proposed controller must compensate for the deficiency angle introduced
by the replacement of the dominant closed loop pole. By following the graphic
258 M. E. Herrera-Cordero et al.

method, the transfer function of the compensated closed loop system is given by

−32.3079(s − 19.088454) K(s + 0.604129)

GT = G(s)s ∗ G(s)comp = ∗
(s + 0.288)(s) s + 1.65520103
(17)
where the calculated gain is K = 2.32967 × 10−3 .
The placement of pole and zeros of the compensated closed loop system is
presented in Fig. 8. The placement of the dominant pole in the desired location
can be observed.

2.5
Root Locus of compenset system
2
System: Root Locus of compenset system
Imaginary Axis (seconds )
-1

Gain: 0
1.5 Pole: -0.5 + 0.866i
Damping: 0.5
Overshoot (%): 16.3
Frequency (rad/s): 1
1

0.5
System: Root Locus of compenset system
0 Gain: 0
Pole: -0.5 - 0.866i
Damping: 0.5
-0.5 Overshoot (%): 16.3
Frequency (rad/s): 1

-1

-1.5

-1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2

-1
Real Axis (seconds )

Fig. 8. Pole and zeros placement in closed loop.

The compensated closed loop system given by Eq. (17) was numerically sim-
ulated in order to appreciate the robot velocity behavior. The obtained results
from this simulation are presented in Fig. 9. In this figure, a comparison between

Fig. 9. Comparison of the linear velocity of the single-wheel pendulum robot with a
constant input torque and with the proposed controller.
 On the Dynamics and Control of a Single-Wheel Robot 259

the open loop system with a constant input torque and the controlled system
is shown. As can be observed, the controller performance presents an overshoot
of about 26% and the settling time is around 205% faster than the open loop
system. Moreover, oscillations are suppressed.
The compensated closed loop system response could be improved by modi-
fying the requirements for the controller. However, it is important to take into
account the physical limitations of the micromotor proposed to drive the single-
wheel robot. In Fig. 10 the required torque for the robot control is shown. As
can be observed, the maximum required torque is 5.325N · mm. This value
can be used to select the micromotor with the adequate characteristics for this
application.

Fig. 10. Required torque for the single-wheel robot control.

5 Conclusions
In this work, the development of the dynamic model of a single-wheel robot with
inertial pendulum locomotion was presented. The obtained mathematical model
was validated by comparing the equations solution with numerical simulations
carried out with the virtual prototype of the robot. The dynamic model was
linearized around an operating point and the linearized model was used to syn-
thesize a closed-loop controller for the forward robot velocity based on the root
locus technique. The designed controller was proved by numerical simulations on
the original non-linear system equations. The closed-loop system response shows
the controller performance achieves the specifications for the robot operation.
At the present time, a first experimental prototype of the single-wheel robot is
being completed in order to experimentally validate the system dynamic model
and the proposed linear controller. The proposed configuration limits the robot’s
movement along a single axis, and as a future work, an extended configuration
to give the robot the capacity of movement in a plane could be proposed.
260 M. E. Herrera-Cordero et al.

References
1. Jin H, Wang T, Yu F, Zhu Y, Zhao J, Lee J (2016) Unicycle robot stabilized by the
effect of gyroscopic precession and its control realization based on centrifugal force
compensation. IEEE/ASME Trans. Mechatron. 21(6):2737–2745. https://doi.org/
10.1109/TMECH.2016.2590020
2. Vos WD, Von Flotow AH, Zhuang H (1990) Dynamics and nonlinear adaptive
control of an autonomous unicycle: theory and experiment. In: Proceedings of the
29th IEEE Conference on Decision and Control, pp. 182–187. IEEE Press, New
York. https://doi.org/10.1109/CDC.1990.203573
3. Jin H, Hwang J, Lee J (2011) A balancing control strategy for a one-wheel pendu-
lum robot based on dynamic model decomposition: simulations and experiments.
IEEE/ASME Trans. Mechatron. 16(4):763–768. https://doi.org/10.1109/TMECH.
2010.2054102
4. Zhu Y, Gao Y, Xu C, Zhao J, Jin H, Lee J (2015) Adaptive control of a gyro-
scopically stabilized pendulum and its application to a single-wheel pendulum
robot. IEEE/ASME Trans. Mechatron. 20(5):2095–2106. https://doi.org/10.1109/
TMECH.2014.2363090
5. Lee J, Han S, Lee J (2013) Decoupled dynamic control for pitch and roll axes of
the unicycle robot. IEEE Trans. Ind. Electron. 60(9):3814–3822. https://doi.org/
10.1109/TIE.2012.2208431
6. Park JH, Jung S (2013) Development and control of a single-wheel robot: practi-
cal mechatronics approach. Mechatronics 23(6):594–606. https://doi.org/10.1016/
j.mechatronics.2013.05.010
7. Lee SD, Jung S (2015) Experimental verification of stability region of balancing a
single-wheel robot: an inverted stick model approach. In: Proceedings of Industrial
Electronics Society, IECON 2015 - 41st Annual Conference of the IEEE, pp. 4556–
4561. IEEE Press, New York. https://doi.org/10.1109/IECON.2015.7392810
8. Herrera-Cordero ME, Arias-Montiel M, Lugo-Gonzalez E (2019) Design and
dynamic modeling of a novel single-wheel pendulum robot. In: Ceccarelli M, Gas-
paretto A (eds) Mechanism Design for Robotics. Springer, Cham, pp 353–360.
https://doi.org/10.1007/978-3-030-00365-4 42
9. Chase R, Pandya A (2012) A review of active mechanical driving principles of
spherical robots. Robotics 1(1):3–23. https://doi.org/10.3390/robotics1010003
10. Roubal J, Husek P, Stecha J (2010) Linearization: students forget the operat-
ing point. IEEE Trans. Educ. 53(3):413–418. https://doi.org/10.1109/TE.2009.
2026427
 Analysis and Simulation of the Oscillations
in a Pulley-Ribbed Belt System

Ernesto Alonso Villegas-Jiménez1(B) , Jorge Díaz-Salgado1 ,

Oliver Marcel Huerta-Chávez1 , Fernando Macedo-Chagolla2 ,
and Edgar Arturo Dávila-Martínez1
1 Tecnológico de Estudios Superiores de Ecatepec, Av. Tecnológico S/N, Col. Valle de
Anáhuac, Sección Fuentes, 55210 Ecatepec de Morelos, Estado de México, Mexico
llanus_2@yahoo.com.mx
2 Facultad de Estudios Superiores Aragón, Av. Hacienda de Rancho Seco S/N, Impulsora

Popular Avícola, 57130 Nezahualcóyotl, Estado de México, Mexico

Abstract. This paper presents the analysis and computer simulation of the rota-
tional and transverse oscillations present in a circular motion transmission sys-
tem through the relationship of pulley-ribbed belt, widely used in the industrial
robots. The analysed system is a pulley-ribbed belt coupled with rigid pulleys. The
analysis is based in the mathematical model developed for serpentine belt drive
systems. The analysis includes: (i) the geometric characteristics, (ii) the behavior
of the transmission in steady state, and (iii) the dynamical response of the system.
Dynamic analysis of these type of systems is complex because: (i) it has coupled
rotational and transverse vibration modes in a non-linear form, and (ii) includes
a circular motion transmission system acceleration-deceleration profile. The pro-
posed analysis allows to determine a priori the frequency, the phase shift and the
oscillation amplitude of a section of the ribbed belt excited by any forced velocity
profile.

Keywords: Dynamic analysis of a pulley-ribbed belt system · Pulley-ribbed belt

system · Rotational modes · Transverse modes

1 Introduction

The rotational motion transmission system based on pulley-ribbed belt coupling, is a

flexible coupled mechanical system with sliding elements, used widely in industrial
robotics, because of the advantages with respect to other types of mechanical couplings.
Although it has been used for more than forty years to transmit mechanical energy
through a circular movement, this system still presents important challenges as: (i) a
detailed mathematical analysis, (ii) the implementation of computer aided design, and
(iii) the study of new applications.
Circular motion transmission systems must operate reliably with minimal noise and
vibration. The problems that normally arise are due to resonance modes that generates

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 261–268, 2020.
https://doi.org/10.1007/978-3-030-45402-9_25
262 E. A. Villegas-Jiménez et al.

noise and fatigue mechanics effects. Since 1979 the belts in “v” have been largely
replaced by serpentine belt drives with dynamic tensioners in transmissions systems [1].
Nowadays, these systems use multichannel ribbed belts, to reduce flexion and
upgrade damper stiffness. Ribbed type transmissions are reliable, reduce the tension
of the belt and the generation of friction heat when operating near the constant tension
of the belt [2].
The use of pulley-ribbed belt system reduces noise and vibration levels. There is
a great interest in the full understanding of the dynamic and static behavior of this
nonlinear mechanical system. Effective tools for the model have been developed over
several years and predict its response. Rotational and vibration transverse modes occur
in all transmission systems that involve belts with fixed pulleys [3].
In rotational modes driven pulleys rotate on their axes and belt sections act as
axial springs. In the transverse modes sections of belt vibrate transversally, similar to a
stretched rope. Both degrade the performance of the system. Rotational modes of sheaves
induce dynamic tension in the belt, dynamic reactions, fatigue effect on the elements
and noise in the structure. The transverse vibrations also induce dynamic tensions and
can radiate directly noise.
Rotational modes in a ribbed belt have been studied during the last twenty five years.
In [4] rotational modes are studied as a stationary system including the support and the
belt damping. In [5] is presented the study of a ribbed belt with a dynamic tensor under
the rapid acceleration of the machine. Finally, in [6] the transient response of a similar
system is examined considering rapid acceleration of the engine, to predict the sliding
of the belt. However, all these studies ignore the transverse dynamics of the belt. In [7]
a dynamic tensioner coupled with a transmission system is studied.
In this work an appropriated mathematical model is proposed. This model must be at
the same time: (i) as simple as possible and (ii) reflect accurately the static and dynamic
behavior of the system. The document is organized as follows: in section number one the
mathematical model is proposed. The computational method used in the static analysis
is described in section two. Results and the respective analysis are presented in section
three. Finally some conclusions are presented.

2 Mathematical Model Deduction

The ribbed pulley-belt system (Fig. 1) is an example of a flexible coupling with sliding
elements, capable of reducing rotational and transverse oscillations. This system was
modeled by [8] under fairly realistic modeling considerations and uses the minimum of
elements to model its behavior which are: (i) a driving pulley, (ii) a driven pulley, (iii) a
dynamic tensor, and (iv) a belt.
Practical considerations for analysis are: (i) the damping factor is considered neg-
ligible, (ii) the belt is modeled with negligible flection stiffness, (iii) the belt speed is
constant and uniform for axial translation, (iv) the belt has an insignificant slip factor.
The transverse vibrations present in the two sections of the belt when attaching a
dynamic tensor are first studied in [8].
 Analysis and Simulation of the Oscillations in a Pulley-Ribbed Belt System 263

Fig. 1. Diagram of the drive system by means of a belt and three pulleys.

Previous research assumes that in the linear response: rotational and transverse move-
ments are decoupled. This approximation is good for systems with fixed elements, but
because the rotation of the tensor arm linearly couples the transverse and rotational
movements, it cannot be used for systems where there is a dynamic tensor.
Hamilton principle is used to derive the non-linear in steady-state equations to
describe the system behavior at constant speed. The nonlinearities considered are geo-
metrical describing a finite stretch of the sections of the belt. The steady-state assessment
is presented in [8] and [9]. The non-linear equation of transverse movement for the three
sections of belt is [10]:
 
m wi,tt + 2cwi,xt − Pti wi,x x =Pdi wi,x x
(i = 1, 2, 3) (1)

Where: c is the belt travel speed constant, i is the belt section number and the subscripts,
x, and t, are used to denote the partial derivative with respect to the coordinate section,
x, and time t, respectively. The 2cwi,xt term is the Coriolis acceleration component.
264 E. A. Villegas-Jiménez et al.

The nomenclature corresponds to:

c = steady state belt speed [m/s]

E A = longitudinal belt stiffness modulus [N /m]
Fdi = dynamic force applied to the pulley i [N ]
Ji = mass moment   of inertia for i-th discrete element, for tensioner arm it is Jar m +
m pulley rar 2 kgm 2
m
k4 = translational stiffness of tensioner arm [N /m]
k gr = tensioner geometric spring stiffness [N /m]
ki = longitudinal stiffness of span i [N /m]
kr = rotational spring constant adimensional of tensioner arm
K s = stiffness rotational of tensioner arm [N /m]
li = length of belt span i [m]
Mdi = dynamic moment applied in the pulley i [N m]
m = belt mass per unit length, constant  throughout
 system [kg]
Pc = balancing centrifugal pressure kg/s 2
 
Pti = traction tension in the spans in equilibrium operation kg/s 2
 
Pdi = dynamic tension in span i-th kg/s 2
ri = i-th system discrete element radius [m]
t = time [s]
u i (xi , t) = longitudinal displacement in the i-th span of the belt [m]
ωi (xi , t) = transversal displacement in the i-th span of the belt [m]
xi = local coordinate in the longitudinal direction of the i-th span of the belt [◦ ]
z i = local coordinate in the transversal direction of the i-th span of the belt [◦ ]
θi = rotation of i-th discrete element in system from equilibrium [◦ ]
ψ1 , ψ2 = alignment angles [◦ ]
ω = frequency of oscillation [s]

Pti is the traction stress in the i-th band element in steady state. The component of cen-
tripetal acceleration c2 wi,xt is included in the term Pti wi,x x . Pulleys motion is described
by the next linear equations [11]:

Pd1 − Pd3 = m 1 ẍ1 (2)

Pd2 − Pd2 = m 2 ẍ2 (3)

Pd3 − Pd3 = m 3 ẍ3 (4)

Where xi = ri θi , m i = Ji /ri2 and Pdi are the dynamic tensions in each section, induced
by an infinitesimal pulley and a rotational arm tensioner.
Well then:

Pd1 = k1 (x3 cos ψ1 + x2 − x1 ) (5)

Pd2 = k2 (x3 cos ψ2 + x4 − x2 ) (6)

 Analysis and Simulation of the Oscillations in a Pulley-Ribbed Belt System 265

Pd3 = Pk3 (x1 − x4 ) (7)

Here ki = E A/(li ), (li ) it is the length of the section i of the band, and 1, 2, are the
angles of alignment between the movement of the tension arm and the adjacent sections
of the band in balance.
The linearized equation of motion of the tension arm is:
 
 t1 ω1,x (l1 , t) + mcω1,t (l1 , t) sin ψ1 . . .
−P
+ Pt2 ω2,x (0, t) − mcω2,t (0, t) sin ψ2 . . .
−k1 (x3 cos ψ1 + x2 − x1 ) cos ψ1 . . . (8)
−k 2 (x3 cosψ2 + x4 − x2 ) cos ψ2 . . .
− ks + k gr x3 = m 3 ẍ3

Where, ks = kr /r32 , kr , are the rotational stiffness of the tensioner spring and

k gr = (Pt1 sin ψ1 − Pt2 sin ψ2 )/r3 (9)

is tensioner arm stiffness when changing your position.

3 Computational Method

A computational method is established for the stationary analysis. The method consists
in determining the roots of the vibration modes function obtained by the modal analysis.
Figure 2 shows the frequency function in steady state regime, this is, with constant
rotational velocity.

Fig. 2. Determinant function vs. Frequency

The function roots can be obtained: (i) in a graphical form or (ii) using an appropriate
numerical method. The proposed algorithm is described next:
266 E. A. Villegas-Jiménez et al.

1. Determine the explicit function for a valid set of parameters.

2. Run an outer loop varying different speed of rotation values within a range of interest.
3. Run an internal loop varying an initial search range determined by user.
4. Using an adequate root search function a vector of founded roots is formed within
the range of interest.
5. Run a last loop to determine the changes between the founded roots, the number of
founded roots and the root chart.

4 Analysis of Results

Figure 3 shows the effect of the variation in the transport speed in the first six oscillation
modes of the system, according to the following parameters:

r1 = 0.0889[m]
r2 = 0.4520[m]
r3 = 0.0970[m]
r4 = 0.0269[m]
l1 = 0.1548[m]
l2 = 0.3449[m]
l3 = 0.5518[m]
ψ1 = 44.21[◦ ]
ψ2 = 178.74[ ◦
 ] 2
J1 = 0.072480kg · m 
J2 = 0.000293kg · m 2 
J3 = 0.001165kg · m 2 
J4 = 0.000293 kg · m 2
kr = 54.37
m = 0.1029[kg]
N
E A = 170 m
 
Pc = 300 kg/s 2

The results show that: (i) the first and fourth mode remains quasi-invariant along the
transport speed variation, (ii) the second mode decrease smoothly as transport speed
increases, (iii) the third, fifth and sixth modes drastically decrease as the transport speed
increases, (iv) the behavior of the first six dominant modes of oscillation can be observed
with respect to the transport speed of the driving pulley using three different sets of angles
for the position of the tension arm (ψ1 and ψ2) and (v) the curves show that the dominant
modes of oscillation are practically independent of the tensioning angles of the tensioner
and dependent on the speed of rotation of the driving pulley. The higher modes are not
included since they are harmonics of the fundamental ones.
 Analysis and Simulation of the Oscillations in a Pulley-Ribbed Belt System 267

Fig. 3. Effect of transport speed in natural frequencies

5 Conclusions

The results obtained in this research show the dependence of the modes of oscillation
as function of the rotational speed of the conductive pulley.
Three algorithms are proposed for the numerical solution of steady state which
allow an analysis of behavior of different system parameters, so that the influence of
each parameter can be seen graphically and quantitative in the oscillation modes of the
system. In particular two types of parametric analysis were proposed: (i) the behavior
of the oscillation modes in the orientation of the tensioner arm and (ii) the behavior of
the first mode of oscillation in the initial tension of the belt sections.
The variation of any of the system parameters affects the modes of vibration, since
they are function of the system eigenvalues, and these, in turn, are a function of the
system parameters.
Although, in principle, the designer can vary freely the values of the parameters, by
the geometric arrangement of the system elements, not any set of parameters is physically
achievable.
Therefore, to determine whether a set of parameters is physically achievable, you
must: (i) construct the system or, (ii) draw the system at scale and determine it graphically.
This process can be angry and time consuming, so it is desirable to have a method that
allows you to carry out the system stroke in a simple way.

References
1. Beikmann RS, Perkins N, Ulsoy A (1996) Free vibration of serpentine belt drive systems. J
Vib Acoust 118(3):406–413
268 E. A. Villegas-Jiménez et al.

2. Casidy RL, Fan SK, Macdonald RS, Samson WF (1979) Serpentine-extended life accessory
drive. In: SAE international, United States
3. Houser DR, Oliver L (1975) Vibration of V-belt drives excited by lateral and torsional inputs.
In: Proceedings of the fourth world congress on the theory of mechanisms, Newcastle Upon
Tyne, England, vol 4, pp 899–905
4. Gaspar RGS, Hawker LE (1989) Resonance frequency prediction of automotive serpentine
belt drive systems by computer modeling. In: Proceedings of ASME conference on mechanical
vibration and noise, vol 18, no 12, pp 13–16
5. Barker CR, Oliver LR, Breig WF (1991) Dynamic analysis of belt drive tension forces during
rapid engine acceleration. In: SAE international, United States
6. Hwang SJ, Perkins NC, Ulsoy AG, Meckstroth R (1994) Rotational response and slip
prediction of serpentine belt drive systems. ASME J Vib Acoust 116(1):71–78
7. Hawker LE (1991) A vibration analysis of automotive serpentine accessory drive systems.
PhD dissertation, University of Windsor, Ontario, Canada
8. Beikmann RS (1992) Static & dynamic behavior of serpentine belt drive systems: theory and
experiment. PhD dissertation, University of Michigan, Ann Arbor, MI
9. Beikmann RS, Perkins NC, Ulsoy AG (1997) Design and analysis of automotive serpentine
belt drive systems for steady state performance. J Mech Des 119(2):162–168
10. Wickert JA, Mote CD Jr (1998) Current research on the vibration and stability of axially-
moving materials. Shock Vib Dig 20(5):3–13
11. Li X, Chen L (2008) Modal analysis of coupled vibration of belt drive systems. Appl Math
Mech 29(1):9–13
 Attitude Control of Torpedo Anchor
with Vectorial Thrust for Use
in a Pseudo-Submersible Platform

G. J. Garnica-Castro1(B) , O. M. Huerta-Chávez2 , J. Cruz-Cruz2 ,

and M. Toledo-Velázquez2
1 Tecnológico de Estudios Superiores de Ecatepec Div. Ing. Aeronáutica, Ecatepec de Morelos,
Estado de México, Mexico
jgibran@tese.edu.mx
2 Instituto Politécnico Nacional, Sección de Estudios de Posgrado ESIME Zacatenco,

Ciudad de México, Mexico

Abstract. In the present days, the hydrocarbons in the ground are limited by
the high human consumptions. Therefore, a necessity of exploring in deep ocean
becomes a reality with huge challenges, one of these challenges is the anchorage
requirement, of course a problem with its installation. The present work deals on
one way to anchorage, the torpedo anchor. This anchorage system works with a
body forces effects added to hydrodynamics effects, due by the interaction between
torpedo’s surface and water. On the body force we have gravity and flotation
(constants), on the other hand, we have a nonconstants hydrodynamics forces
that can change the trajectory during the fall, and producing a non-successful
installation of the torpedo on the seabed. So the dynamics of the torpedo anchor
are derived by doing a rigid body analysis, computing the general equations that
describe the torpedo trajectory, second, performs a CFD analysis to obtain the
hydrodynamics forces, and by the last, add on an extra vectoring control force (to
correct the orientation).

Keywords: Computational fluid dynamics · Deep penetrating anchor · RANS ·

Deep ocean platform

1 Introduction

For the anchorage presses, the current system of the torpedo is under the action of the
gravity force and the active changes in the hydrodynamics proprieties, this must be
analyzed to guarantee the penetration on the seabed.
This work includes a dynamic analysis of Deep Penetration Anchor (DPA), added
a vectorial “control” force to correct the trajectory. So, it’s necessary perform a mathe-
matical analysis to obtain a mathematical model for describe the dynamical trajectory
in function of the gravity and hydrodynamic proprieties.

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 269–278, 2020.
https://doi.org/10.1007/978-3-030-45402-9_26
270 G. J. Garnica-Castro et al.

1.1 Mathematical Model

As we know, the full description of any 3D rigid body has 6 Degrees of Freedom, so,
it’s necessary obtain the 6 equations that describe the dynamic motion of the torpedo.
According to [1], the vectorial form of this equations is:

M v̇ + C(v)v + D(v)v + g(η1 ) = τ + w (1)

η̇ = J (η2 )v (2)

y =η+u (3)

Where η is the earth-fixed position vector (that describe, of course, the attitude), v
the body-fixed vector of linear and angular velocities, τ are the vector control forces, and
w the disturbance vector. Therefore, the DPA motion is defined in the body-fixed frame
on relation to the earth-fixed frame (considered as inertial frame). Thus, its necessary do
a relation between both frames. The relation of these frames is show in the next Fig. 1.

Fig. 1. Torpedo’s frame of reference related to the earth fixed frame.

Do [2] proposes to divide the analysis in its kinematic and dynamic parts.
For the kinematic analysis, we use the Eq. (3), where η2 is the angle vector
defined as η2 = [φ,θ , ψ] T , so
 i 
Rb (η2 ) 03x3
J (η2 ) = (4)
03x3 T −1 (η2 )
⎡ ⎤
1 sin ϕ tan θ cos ϕ tan θ
T −1 (η2 ) = ⎣ 0 cos ϕ − sin ϕ ⎦ (5)
0 sin ϕ sec θ cos ϕ sec θ
 Attitude Control of Torpedo Anchor with Vectorial Thrust 271

⎡ ⎤
cos ϕ cos θ cos θ sin ψ − sin θ
⎢ ⎥
Rbi (η2 ) = ⎣ − cos ϕ sin ψ + cos ϕ sin θ sin ψ cos ϕ cos ψ + sin ϕ sin θ sin ψ cos ϕ sin θ ⎦
cos ϕ sin θ sin ψ + sin ϕ sin ψ cos ϕ sin θ sin ψ − cos ϕ sin ψ cos ϕ sin θ
(6)
As we know, the dynamical analysis follows the Newton’s second law

Fbi = m(v̇1 + v2 vbi ) (7)

Mbi = m(h ib + v2 h ib ) (8)

T T
Where v1 = u v w , v2 = p q r y h = I v2 , and I is the identity matrix.

1.2 Thrust Force

The following diagram shows the thrust force generated by the nozzle in the body-fixed
frame, noted by F p (Fig. 2).

Fig. 2. Free body diagram with the thrust force acting on the torpedo.

The force and moment by the thrust are

F p = F px î + F py ĵ + F pz k̂ (9)

M p = M px î + M py ĵ + 0k̂ (10)

Of course, the components are the projections in each axis, i.e. F px = F p cos θ p
cos ϕ p , F py = F p cos θ p sin ϕ p F pz = F p sin θ p , M px = (F p cos θ p cos ϕ p )(l p ) and
M py = (F p cos θ p sin ϕ p )(l p ), where l p is the distance.
272 G. J. Garnica-Castro et al.

1.3 Hydrodynamics Perturbations

The forces and moments by the hydrodynamics perturbations are described in the Fig. 3.

Fig. 3. Hydrodynamics forces

Therefore, the forces of hydrodynamic perturbations are

FL x = −q∞ C L A cos ψh (11)

FL y = −q∞ C L A sin ψh (12)

FD = −q∞ C D A (13)

Where C L and C D are the hydrodynamic coefficients (that are obtained by a CFD
analysis), and q∞ is the dynamic pressure, which is defined by q∞ = 21 ρ(vhb )2 .

2 CAD-CFD

The analysis is performed in a conventional torpedo anchor used by Petrobras in Brazil

[3], shown in the Fig. 4, and its characteristics are summarized in Table 1.
We perform a classical CFD approach, hence the control volumen and mesh must
be in accordance to the problem.
The control volumen is in a pastel shape, this based on that the above equations are
derived in the body-fixed axis, so the calculated hidrodynamics forces does not need to
be transformed. This control volume is indicated in the Fig. 5.
In this figure, we can observe the inlet velocity at different angles, this can be
performed easily in a pastel control volumen.
 Attitude Control of Torpedo Anchor with Vectorial Thrust 273

Fig. 4. Isometric view of the torpedo model.

Table 1. Physical Propeties of the torpedo

Volume 0.0086 m3
Area 0.5547 m2
Mass 67.4385 kg
Weight 661.3465 N

Fig. 5. Control volumen of the torpedo at differents angles.

The mesh of the body was performed in professional software, considering an Ortho-
Trias elements. This mesh is specified in the next figure, where we can see a densification
in the fins (Fig. 6).
To detect the viscous effects on the body surface, that has a high impact on the
tangential drag forces, its necessary increased the elements close to surfaces. This process
is summarize to compute the boundary layer thickness
0.377
δ= √3
(11)
Re
274 G. J. Garnica-Castro et al.

Fig. 6. Torpedo mesh view

Where Re is the Reynold’s number. After, the boundary layer is divided by ten layers.
The resulting volumetric mesh used in this analysis is shown in the Fig. 7.

Fig. 7. Side view of the volumetric mesh

The Navier-Stokes equations are simplified to its incompressible form and solved
in a classical Reynolds Averaged Navier-Stokes (RANS) coupling the moment and
continuity equations [4]. The turbulent model is the Spalart Almaras (S-A), with one
equation for the turbulent viscosity including the convective and transport elements to
calculate the production and dissipation of turbulent viscosity [5].
For the torpedo, the Fig. 8 shows the contours of the velocity and pressure fields as
a result of the numerical simulations.
 Attitude Control of Torpedo Anchor with Vectorial Thrust 275

Fig. 8. Velocity and pressure contours

This figure shows the instabilities of the fluid at angles greater of 5, this impact
directly on the stability of the torpedo. To show this behavior we made a graph of the
moment coefficient and angle of attack (Fig. 9).
We can see that the torpedo has a high stability in angles of attack between −5 to +
5. With these results we can continue with the control.
276 G. J. Garnica-Castro et al.

Fig. 9. Moment coefficient vs angle of attack

3 Control Design

The equations that describe the orientation of the torpedo are

⎡ ⎤ ⎡ ⎤ ⎡ ⎤
M px Mhx I x ṗ − (I y − Iz )qr
⎣ M py ⎦ + ⎣ Mhy ⎦ = ⎣ I y q̇ − (Iz − I x )r p ⎦ (12)
0 0 Iz ṙ − (I x − I y ) pq
⎡ ⎤ ⎡ ⎤
I x [(I y− Iz )qr + Mhx + M px ] ⎥
1
⎢ ṗ ⎥ ⎢
⎢ ⎥ ⎢ ⎥
⎢ ⎥ ⎢ ⎥
⎢ q̇ ⎥ = ⎢ 1
[(I − I )r p + M + M py ⎥
(13)
⎢ ⎥ ⎢ Iy z x hy ⎥
⎣ ⎦ ⎣ ⎦
ṙ 1
Iz [(I x − I y ) pq

Or

ẋ = f (x) + g(π )
(14)
y = h(x)

Where f(x) is the state function and g(π ) the input function. Hence the equation that
describe the output of the system in terms of the perturbation and propulsion force are
⎡ ⎤ ⎡ ⎤
⎢ qr (I x − Iz ) + Ix Mhx ⎥ ⎢ Ix (F p cos θ p cos ϕ p )(l p ) ⎥
1 1
[ ÿ] = ⎣ ⎦+⎣ ⎦ (15)
r p(Iz − I y ) + I1y Mhy 1
Iy (F p cos θ p sin ϕ p )(l p )

And

s2 ṡṡ = −|s2 |D2 − s2 d2 ≤ 0 (16)

Using the input parameters obtained in the CFD analysis we can simulate the
behaviour of the torpedo at different angles of attack (Figs. 10 and 11).
Attitude Control of Torpedo Anchor with Vectorial Thrust 277

Fig. 10. Torpedo at 4 degrees.

Fig. 11. Torpedo at 8 degrees

278 G. J. Garnica-Castro et al.

4 Conclusions
In this work, we obtain the dynamic equations that describe the motion of a DPA with
a thrust, limiting only to a dynamic analysis. These equations are the first approach
to obtain a control model to modify the trajectory. The CFD analysis gives a more
precise data to complete the dynamic equations, the next step is implementing a control
technique, and the design the devise to give the thrust force.

References
1. Fossen T (2011) Handbook of marine craft hydrodynamics and motion control. Wiley
2. Khac Duc Do JP (2009) Control of ships and underwater vehicles. Springer-Verlag London
3. Fernandes C (2011) Directional stability of the torpedo anchor pile during its installation. IES
J Part A Civil Struct Eng 4:180–189
4. Wilcox D (1994) Turbulence modeling for CFD. DCW Industries, Inc
5. Alder B (1964) Methods in computational physics: advances in research and applications.
Academic Press
6. Kermode A (2006) Mechanics of flight. Pearson
7. Renilson M (2015) Submarine hydrodynamics, Springer briefs in applied sciences and
technology
8. Antonelli G (2006) Underwater robots, Springer tracts in advanced robotics
9. Stevens B (2003) Aircraft Control and Simulation. Wiley-Interscience Publication
10. Do D (2009) Control of ships and underwater vehicles. Springer
 Geometric Parameters and Workspace
Optimization of Sitting-Type Lower Limb
Rehabilitation Robot

Askhat Diveev1 , Grigory Dubrovin2 , Dmitry Malyshev3(B) ,

and Anna Nozdracheva4
1 Federal Research Center “Computer Science and Control” of Russian Academy of Sciences,
Moscow, Russia
2 Kursk State Medical University, Kursk, Russia
3 Belgorod State Technological University named after V.G. Shukhov, Belgorod, Russia
rlbgtu@gmail.com
4 Gamaleya National Research Center for Epidemiology and Microbiology, Moscow, Russia

Abstract. A lower limb rehabilitation system based on a serial RRR manipulator

and an active parallel 2PRP-2PPR manipulator is considered, effective numerical
methods and algorithms are developed and tested that allow determining the opti-
mal geometric parameters of the active manipulator mechanism that provides the
required movements within the workspace. The influence of the range of rotation
angles of the joints, which must be ensured during the rehabilitation process on
the minimum overall dimensions of the 2PRP-2PPR manipulator, is analyzed.

Keywords: Workspace · Lower limb rehabilitation · Parallel robot ·

Optimization

1 Introduction
Worldwide, limb disability is one of the major problems faced in the medical sector
and it is more common in the elderly age group. There are numerous reasons for these
disabilities; one major reason is stroke which happens due to brain attack, spinal cord
injury (SCI) which happens due to accidental injury to the spine, disability due to non-
polio acute flaccid paralysis (NPAFP), etc. In addition, stroke is one of the common
causes of mortality after coronary artery disease and it is the commonest cause of chronic
adult disability as well. The lifetime risk of stroke after 55 years of age is 1 in 5 for women
and 1 in 6 for men [1]. Besides, another kind of disease which is acute flaccid paralysis is
common. Clinically indistinguishable from polio paralysis but acute flaccid paralysis is
twice as deadly as polio paralysis and takes longer duration to recover from this disease
[2]. These limb impairments need to go through rehabilitation process to regain normal
functioning of limb activities. However, rehabilitation therapeutic treatments are limb
specific and they are different from upper limbs to lower limbs. Upper limb treatments
focus on regain of hand neural, muscular action ability and strength while lower limb
treatments focus on various joint movements of the legs and their synchronized motions.
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 279–289, 2020.
https://doi.org/10.1007/978-3-030-45402-9_27
280 A. Diveev et al.

Lower limb rehabilitation or therapeutic treatment has been a warm topic in recent
years as mechanized systems have promised effective results and produced significant
improvements in patients’ recovery through the robotic physiotherapy [3]. In the lower
limb treatments: every lower limb has a specific movement (body and limb specific)
and it requires a specialized setup for the individual limb. These issues make the system
development quite complex and require modular and robust mechanical systems for
the lower limb treatments. However, lower limb treatments are very common in the
rehabilitation domain and cover a wide range of patients.
It is widely accepted that human locomotion depends both on basic patterns generated
at the spinal level, and the prediction and reflex-dependent fine control of these patterns
at different levels [4]. These physiological motions recorded from healthy subjects are
implemented on lower-limb impairment patients. These data sequence for lower-limb
joints are called gait data (walking pattern). Other than gait data lower-limbs have move-
ments like hip-flexion and extension, knee flexion and extension, ankle plantar flexion
and dorsiflexion. Gait training rebuilds muscular action synchronization in the lower-
limb and for strengthening each leg-joint is treated separately for regaining the motor
actions.
For the rehabilitation of the lower limbs, mechanotherapy is widely used. For this
purpose, such commercially available rehabilitation under development. Most rehabili-
tation systems have an active orthosis in their design that has a simple structure, but their
dynamic characteristics are low due to the presence of drives in the lower limb orthosis
system. Other types of rehabilitation systems have linear drives and high dynamic char-
acteristics, but at the same time, they have a limited workspace with limited movement of
linear drives or do not include supporting or orthopedic systems [5, 6]. The development
of mechanisms for the rehabilitation of the lower limbs is also considered in the works
[7–9]. An important task is the design of mechanisms with enhanced structural rigidity
of the mechanical structure, ergonomics, compactness of the system, positioning accu-
racy, improved weight and size characteristics and reduced dynamic load on the drives
to ensure stable and safe movement of the lower and upper limbs of patients, including
using parallel mechanisms that are well studied in [10–12].

2 Formulation of the Problem

In this paper, a sitting/lying type lower limb rehabilitation robot is considered for the
performance analysis which is providing an improved rehabilitation robotic system. The
conceptual design of the system is given in Fig. 1. The device consists of two mechanisms
namely a planar 2PRP-2PPR parallel manipulator and a planar RRR serial manipulator
as a passive orthosis system.
Consider the system (Fig. 2), where in addition to the passive orthosis in the form
of a four-link RRR mechanism, the active 2PRP-2PPR mechanism is used. An RRR
mechanism is used to secure a person’s legs. Its revolute joints of the RRR mechanism
correspond to the patient’s joints: P1 - to the patient’s hip joint, its angle of rotation
relative to the horizontal axis X is indicated by α, the joint P2 with the angle of rotation
γ relative to the 1 link and β relative to the horizontal axis X corresponds to the knee
joint, and P3 with the angle rotation θ relative to the beam perpendicular to the link
2 -ankle joint.
 Geometric Parameters and Workspace Optimization 281

Fig. 1. Conceptual design of the lower limb rehabilitation system

The active 2PRP-2PPR mechanism provides the rehabilitation angles of rotation of

all joints of the patient and coordinates x P3 , z P3 of the ankle joint P3 of the patient in
the X Z plane required for rehabilitation. In the designation 2PRP-2PPR, the letter P
denotes prismatic joints, R - revolute joints, numerals 2 - two similar kinematic chains.
Links 1, 3 and 4 move progressively relative to the fixed base 0. Link 2 rotates relative
to link 1 due to the revolute joint A3 . Link 5 moves along the Z axis and along link 2,
and also rotates relative to link 4 due to the revolute joint A4 . Link 6 moves along link
2. Link 8 is connected to link 6 through a gear reducer with gear ratio i 12 = ωω21 = 0, 25.
The article described the task of determining the optimal geometric parameters of
a parallel 2PRP-2PPR mechanism that provides movement of a passive orthosis - RRR
mechanism within the workspace in accordance with the required joint movements.

Fig. 2. The design scheme of the rehabilitation mechanism.

282 A. Diveev et al.

3 Kinematics Analysis
Consider the based kinematic relations of manipulators. The end-effector position is
obtained from the forward kinematic relation of the parallel manipulator. The forward
kinematic relation is given in equation:
 T
μ = x P3 z P3 ψ = f (η) (1)
     T
μ = x A2 z A3 + (x A3 −x A2 )(z A4 −z A3 )
(x A3 −x A4 ) + L p tan −1 (z A4 −z A3 )
(x A3 −x A4 ) (2)

where, μ p ∈ R3×1 is the vector of operational-space positions, i.e., end-effector posi-

tions and orientation. η ∈ R3×1 is the vector of linear displacements of the actuated linear
 T
joints i.e., η = x A2 z A3 z A4 (configuration or joint space positions or displacements).
We express z A3 and z A4 from formulas (1)–(2):

z A4 = z P3 − L p + (x P3 − x A4 ) tan ψ (3)

z A3 = z P3 − L p − (x A3 − x P3 ) tan ψ (4)

Similarly, the end- effector position is obtained from the forward kinematic relation
of the serial manipulator (i.e., a passive orthosis) and is given in equation:
 T
μs = x z ψ = f () (5)
⎡ ⎤
x1 − L thing cos α − L cr us cos β
μs = ⎣ y1 + L thing sin α + L cr us sin β + L p ⎦ (6)
G(α + β + θ)

where  ∈ R3×1 is the vector of angular displacements of the passive joints of the
 T
orthosis i.e.,  = α β θ .
For designing a rehabilitation system to ensure its compactness, it is important to
determine the minimum possible overall dimensions at which it will perform its functions
with the given parameters. In this case, these parameters are the sizes of the lower limbs
of the patient and the range of rotation angles of the joints required for rehabilitation.
Determination of the RRR mechanism workspace is used to calculate the optimal size
of the 2PRP-2PPR mechanism.

4 Workspace Definition

Consider the problem of determining the boundaries of the workspace of the RRR mech-
anism when changing the parameters of the mechanism: angles α and γ in accordance
with the movements of the passive orthosis, which is modeled by this mechanism. For
rehabilitation, it is necessary to provide the following angles in the hip, knee and ankle
joint when simulating walking: 1 step phase - flexion in the hip joint from 0° to 20°,
 Geometric Parameters and Workspace Optimization 283

flexion in the knee joint from 0° to 60°, extension in the ankle joint from 0° up to 10°; 2
phase - 0°, 0° and 90° respectively; 3 phase - extension in the hip joint from 0° to 10°,
in the knee joint °, flexion in the ankle joint from 0° to 20°.
To determine the workspace, we will use the method of non-uniform coverings, which
is described in detail in [13–15] for other types of parallel robots. For this purpose, we
will use the methods of interval analysis described in detail in [16, 17].
We record the intervals that describe the ranges of changes in the rotation angles of
the patient’s joints. The lower boundary of the intervals corresponds to the minimum
values of the angles necessary for rehabilitation, and the upper maximum:
 
A := A,  = A ≤ α ≤ A
 
 := ,  =  ≤ γ ≤ 
 
 := ,  =  ≤ θ ≤ 
 
β = α + γ , that is, B := B, B = A +  ≤ β ≤ A + 
To reduce the dimension of the computational problem, we propose using the
following approach:
We express the angle α from (5)–(6):
z P3 − L cr us sin β
α = sin−1
L thigh

Denote: f 1 (z P3 , β) = sin−1 z P3 −L cr us sin β

L thigh , f 2 (a, β) = −L thigh cos α − L cr us cos β.
Consider the intervals
 of values
 guaranteed to include the ranges
 of admissi-
ble values Z P3 : = Z P3 , Z P3 = Z P3 ≤ z P3 ≤ Z P3 , X P3 : = X P3 , X P3 =
X P3 ≤ x P3 ≤ X P3 and the interval B.
For the interval Z P3 and the interval B we define the interval A of the values of the
angle α:
 
A := A , A , where (7)

A = minz P3 ∈Z P3 ,β∈B f 1 (z P3 , β), A = maxz P3 ∈Z P3 ,β∈B f 1 (z P3 , β)

For the interval A and the interval B we define the interval   of the values of the
angle γ :
  = B − A (8)
 
Then, for the intervals A and B, we calculate the interval X  P3 := X  P3 , X  P3 ,
where
X  P3 = mina∈A ,β∈B f 2 (a, β), X  P3 = maxa∈A ,β∈B f 2 (a, β)
If the system is running
⎧
⎨ A ∩ A = ∅
X ∩ X  P3 = ∅ (9)
⎩ P3
 ∩   = ∅
284 A. Diveev et al.

then for intervals Z P3 , X P3 and B there is at least one point entering the workspace.
The interval constraints of the variables form an n-dimensional box, where n corre-
sponds to the number of variables, i.e. n = 3. To determine the workspace, the initial
and subsequent boxes, for which (9) holds, are divided into smaller boxes. For each box
B obtained as a result of the partition, the verification is performed similarly. The work
area is described by the set P I of boxes B, for which d(Pi ) ≤ δ, where δ is a given pos-
itive value that determines the accuracy of the approximation. For interval calculations,
we use the Snowgoose interval analysis library [18]. Next, we determine the optimal
geometric parameters of the 2PRP-2PPR mechanism.

5 Optimization of Geometric Parameters of the 2PRP-2PPR

Mechanism

Consider the task of determining the optimal parameters of the 2PRP-2PPR mechanism,
providing the movement of the output link of the RRR mechanism within a given work
area.
Using a set of boxes that describe the workspace, we will choose the optimal overall
dimensions along the Z axis:

x A3 = max B∈P I X P3 (10)

x A4 = min B∈P I X P3 (11)

We determine the optimal angle ϕ between the link P3 P4 of the RRR mechanism
and the link A3 A4 of the 2PRP-2PPR mechanism with the coordinates z A3 = z A4 of
the joints A3 and A4 equal. As an optimization criterion, we choose a function that
determines the distance h along the Z axis between the hinges A1 and A2 , equal to the
distance between the minimum and maximum values of z A3 and z A4 . Using formulas
(3)–(4), we obtain:

f 3 (ϕ) = min h
ϕ∈0,..,90◦
= min max B∈P I maxi∈1,..,4 gi − min B∈P I mini∈1,..,4 gi , where (12)
ϕ∈0,..,90◦
 
Z P3 + Z P3 X P3 + X P3
g1 = − L p + t1 − x A4 ,
2 2
 
Z P3 + Z P3 X P3 + X P3
g2 = − L p + t2 − x A4 ,
2 2
 
Z P3 + Z P3 X P3 + X P3
g3 = − L p − t1 x A3 − ,
2 2
 
Z P3 + Z P3 X P3 + X P3
g4 = − L p − t2 x A3 − ,
2 2
 Geometric Parameters and Workspace Optimization 285

where
  
B+B π
t1 = tan 0, 25 + +−ϕ
2 2
  
B+B π
t2 = tan 0, 25 + +−ϕ
2 2

The values of the functions gi are determined for each of the boxes of the list P I .
We define the minimum and maximum values of the coordinates of the joints A1 and
A2 along the Z axis for the set of boxes of the list P I obtained earlier and the angle ϕ.
This procedure is performed for angles ϕ ∈ [0◦ , 90◦ ] with a step of 1◦ . The angle ϕ at
which the distance h has a minimum value is assumed to be optimal:

h = max max gi − min min gi

B∈P I i∈1,..,4 B∈P I i∈1,..,4

Using these data, we will choose the optimal overall dimensions of the 2PRP-2PPR
mechanism along the Z axis, which are determined by the following relationships:

z A1 = max B∈P I maxi∈1,..,4 gi (13)

z A2 = min B∈P I mini∈1,..,4 gi (14)

6 Algorithm Synthesis
To solve the tasks of determining the workspace and choosing the optimal geometric
relationships of the parallel 2PRP-2PPR mechanism, an algorithm is synthesized that
works with two lists of three-dimensional boxes P and P I . The block diagram of the
algorithm is shown in Fig. 3.
The algorithm works as follows:

1. Set the geometric parameters of the robot, the intervals A, ,  and the accuracy
of the approximation δ.
2. At the first step of the algorithm, the list of internal approximations P I is empty,
the list P consists of only
  one box Q, including   the intervals Z P3 , XP3 and
B: Z P3 = X P3 = − L thigh + L cr us + L p , L thigh + L cr us + L p , B =
 
A + , A +  .
3. Remove the box B from the list P.
4. If d(B) < δ, then add B to the list P I and go to step 13.
5. Calculate A by the formula (7).
6. If A > 1 or A < −1, then go to step 13.
7. Calculate   by the formula (8).
8. If A ∩ A = ∅ or  ∩   = ∅, then go to step 13.
9. Calculate X  P3 .
286 A. Diveev et al.

Fig. 3. The block diagram of the algorithm for determining the workspace and optimization of
geometric parameters.

10. If X P3 ∩ X  P3 = ∅, then B is divided into two equal boxes along the edge with
the longest length. Insert these boxes at the end of the list P.
11. If P = ∅, then go to step 3.
12. Calculate x A3 and x A4 using formulas (10)–(11).
13. Calculate f 3 by the formula (12). The angle ϕ, for which h is minimal, is optimal.
14. Calculate z A1 by formula (13) and z A2 by formula (14).

To implement the synthesized algorithm, a program was developed in the C++ pro-
gramming language with the ability to output results for visualization in the Gnuplot
program.
 Geometric Parameters and Workspace Optimization 287

7 Simulation Results
In accordance with the parameters required for rehabilitation: A = [−20◦ , 10◦ ],
 = [−60◦ , 0◦ ],  = [−10◦ , 20◦ ], B = [−80◦ , 10◦ ], L thigh = 450 mm, L cr us =
450 mm, L P = 10 mm, a computational experiment was conducted, the simulation
results are presented in Fig. 4 and 5. The calculation time for the approximation accu-
racy δ = 1 mm on a personal computer was 27 s, of which 21 s were modeling the
workspace, 6 s were optimizing the geometric parameters of the 2PRP-2PPR mecha-
nism and determining the angle ϕ. It can be seen from Fig. 4 that the overall dimensions
of the workspace are 401 × 754 mm. The optimal angle ϕ = 0◦ , the coordinates of the
joints characterizing the overall dimensions and position of the 2PRP-2PPR mechanism
were determined: x A3 = −500 mm, x A4 = −901 mm, z A1 = 164 mm, z A2 = −598
mm. In Fig. 5a shows the position of the joints of the 2PRP-2PPR mechanism, in which,
to ensure the required positions of the RRR mechanism, the maximum value of the
coordinate z A4 is reached at the optimum angle ϕ. In Fig. 5b shows the position of the
mechanisms for the minimum value of the coordinate z A3 .

Fig. 4. The workspace of the RRR mechanism.

Note that the angle ϕ has a different value depending on the required rotation angles
of the patient’s joints. Defined and shown in Fig. 6 the dependence of the distance h on
the angle ϕ at various intervals of the angle of rotation of the patient’s hip joint. For each
of the intervals, the angle ϕ was determined at which the distance h is minimal. For the
interval A = [−20◦ , 10◦ ] the optimal angle is ϕ = 0◦ , for A = [−10◦ , 20◦ ] the angle is
ϕ = 10◦ , for A = [0◦ , 30◦ ] angle ϕ = 21◦ .

8 Conclusion
Thus, for the proposed lower limb rehabilitation system based on a serial RRR manip-
ulator and an active parallel 2PRP-2PPR manipulator, effective numerical methods and
algorithms were developed and tested that allowed us to determine the optimal geomet-
ric parameters of the mechanism of the active manipulator that provides the required
288 A. Diveev et al.

Fig. 5 The simulation results a - the extreme position of the joint A4 for ϕ, b - the extreme position
of the joint A3 for ϕ.

Fig. 6. Dependence of the distance h on the angle ϕ at various intervals of the angle of rotation
of the patient’s hip joint.

movements within the workspace. The calculation time for the approximation accuracy
δ = 1 mm on a personal computer was 27 s. In the simulation, the joints were considered
as material points, and for a real model of the mechanism, the overall dimensions of the
2PRP-2PPR mechanism should be increased taking into account the actual dimensions
of the drive elements to exclude their collision.

Acknowledgements. This work was supported by the Russian Science Foundation, the agreement
number 19-19-00692.

References
1. Seshadri S, Beiser A, Kelly-Hayes M, Kase CS, Au R, Kannel WB (2006) The lifetime risk
of stroke: estimates from the Framinham study. Stroke 37:345–350
2. Vashisht N, Puliyel J (2012) Polio programme: let us declare victory and move on. Indian J
Med Ethics 9(2):114–117
3. Truelsen T, Bonita R (2009) The worldwide burden of stroke: current status and future
projections. Handb Clin Neurol 92(3):327–336
 Geometric Parameters and Workspace Optimization 289

4. Loeb GE (1989) Neural control of locomotion. BioSciences 39:800–804

5. Mirbagheri MM, Tsao C, Pelosin E (2005) Therapeutic effects of robotic-assisted loco-
motor training on neuromuscular properties. In: Proceedings of the IEEE 9th international
conference on rehabilitation robotics (ICORR), Chicago, USA, pp 561–564
6. Hachisuka K (2010) Robot-aided training in rehabilitation IIJ. Brain Nerve 2:133–140
7. Hsieh CC, Kuo CH, Matsuura D, Takeda Y (2019) A 7R spatial linkage for ankle rehabilitation
with an arbitrary ankle rotation axis. In: IFToMM world congress on mechanism and machine
science, pp 1491–1498
8. Zhang C, Lan B, Matsuura D, Mougenot C, Sugahara Y, Takeda Y (2018) Kinematic design
of a footplate drive mechanism using a 3-DOF parallel mechanism for walking rehabilitation
device. J Adv Mech Des Syst Manuf 12:JAMDSM0017
9. Rios A, Hernandez E, Moreno JA, Keshtkar S, De la Garza R (2018) Kinematics analysis of a
new 3DOF parallel manipulator as walking rehabilitation device a rios. In: 15th international
conference on electrical engineering, computing science and automatic control (CCE)
10. Aleshin AK, Glazunov VA, Rashoyan GV, Shai O (2016) Analysis of kinematic screws that
determine the topology of singular zones of parallel-structure robots. J Mach Manufact Reliab
45(4):291–296
11. Kong H, Gosselin CM (2007) Type synthesis of parallel mechanisms. Springer, Heidelberg
12. Rybak LA, Khalapyan SY, Gaponenko EV (2018) Issues of planning trajectory of parallel
robots taking into account zones of singularity. IOP Conf Ser Mater Sci Eng 327:1–8 042092
13. Evtushenko YuG, Posypkin MA, Rybak LA, Turkin AV (2017) Finding sets of solutions to
systems of nonlinear inequalities. Comput Math Math Phys 57(8):1241–1247
14. Evtushenko Yu, Posypkin M, Turkin A, Rybak L (2017) The non-uniform covering approach
to manipulator workspace assessment. In: Proceedings of the 2017 IEEE Russia section young
researchers in electrical and electronic engineering conference, ElConRus, pp 386–389. IEEE
15. Evtushenko Y, Posypkin M, Rybak L, Turkin A (2018) Approximating a solution set of
non-linear inequalities. J Global Optim 7:129–145
16. Jaulin L (2001) Applied interval analysis: with examples in parameter and state estimation,
robust control and robotics. Springer, New York
17. Moore RE (1966) Interval analysis. Prentice-Hall, Englewood Cliffs
18. Posypkin M, Usov A (2019) Basic numerical routines. https://github.com/mposypkin/
snowgoose. Accessed 18 Nov 2019
 Comparison of Parallel Versions
of SA and GA for Optimizing
the Performance of a Robotic
Manipulator

Adán H. Baena1,3(B) , Sergio I. Valdez2,3(B) ,

Felipe de Jesús Trujillo Romero1,3(B) , and Moisés M. Montes1,3(B)
1
División de Ingenierı́as, Universidad de Guanajuato,
Carr. Salamanca - Valle de Santiago km 3.5 + 1.8,
Palo Blanco, 36885 Salamanca, GTO, Mexico
{a.hernandezbaena,fdj.trujillo,mm.montesvillanueva}@ugto.mx
2
Centro de Investigación en Ciencias de Información Geoespacial,
CENTROGEO A.C., Contoy 137, Lomas de Padierna,
14240 Ciudad de México, Mexico
sergio.valdez@conacyt.mx
3
Cátedra CONACYT, Ciudad de México, Mexico

Abstract. The objective of this work is to optimize a robotic manipu-

lator with five links and two degrees of freedom. Our optimization crite-
rion is the lowest energy consumption subject to delivering an adequate
performance, this means that the mechanism should follow the defined
trajectory with a maximum bounded error. For this purpose, we used the
Simulated annealing (SA). which is a probabilistic technique for approx-
imating the global optimum of a given function. In our case, the given
function is the integral of the torque returned by the simulation of the
mechanism. The SA evaluates the function with random configurations
and according to the returned value, the configuration is accepted or
rejected in agreement with an acceptance criterion. The stopping cri-
terion is that three consecutive optimization loops do not improve the
current best solution. Energy consumption is directly proportional to
the torques signals τ 1 and τ 2, so we propose a criterion which considers
energy consumption as well as the position error, as follows:

n 
n
E(l1 , l2 , l4 , k1p , k2p , k1d , k2d ) = λ (τ1i + τ2i ) + (ep1 + ep2 ) (1)
i i

Where τ1 is the torque of link l1 and τ2 is the torque of link l2 as it is

shown in Fig. 1, λ is a penalty value that indicates the importance of the
energy with respect to the position error. It should be mentioned that
the larger the λ, the greater tracking error in the trajectory. The second
part of Eq. (1) indicates the numerical integral of position error, where
ep1 is the error caused by the angle ϕ1 and ep2 is the error caused by

S. Ivvan Valdez is supported by grant Cátedra CONACYT 7795.

c The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 290–303, 2020.
https://doi.org/10.1007/978-3-030-45402-9_28
 Comparison of Parallel Versions of SA and GA 291

the angle ϕ2 . The SA step with the greatest computational cost is the
simulation and evaluation of candidate configurations since each of them
is independent of each other, this part is highly parallelizable, that is
to say, several configurations are evaluated in parallel. Each one of the
threads simulates a configuration with its own memory and only share
specific information, such as the best configuration found so far. The
results demonstrate the time reduction as well as the high quality of the
results. In addition, we compare two versions of Parallel SA and a sim-
ple GA. The first SA is completely equivalent to the sequential algorithm
but the computing time is reduced by parallelizing a set of simultane-
ous evaluations, even though the synchronized acceptance/rejection step
introduces overhead in the algorithm. On the other hand, the second SA
eliminates the synchronization, so it is not equivalent to the sequential
version, but in contrast, it the overhead is reduced. Finally, both versions
are contrasted with a simple GA, in the objective function of the best
solution as well as in computational time.

Fig. 1. Robotic manipulator

Keywords: Parallel · Optimization · Simulated annealing · Genetic

Algorithm

1 Introduction
In all areas of engineering, optimization is a way of improving designs, it has
served as a tool to save material, time, effort, energy, etc. When working in
a new design, mechanical engineers need to analyze and test each candidate
configuration, usually, in a test-error cycle they improve an initial design with
respect to a given criterion. To test a specific configuration of a mechanism, a
computer simulation is convenient in order to save resources, and reducing costs
and time. In addition, an adequate computer program could test much more
candidate configurations than a human designer, nevertheless, it must do it in
an adequate an efficient way.
292 A. H. Baena et al.

The Simulated Annealing Algorithm (SA) [1] is a global optimization algo-

rithm based on Gibbs probability distribution. It generates random configura-
tions from the Gibbs distribution using the metropolis sampling method, the
better a configuration is the greater the probability to be sampled. Nevertheless,
a drawback of the SA is the need of a huge amount of objective function evalu-
ations, in this case an evaluation is a numerical simulation of a particular robot
configuration. The SA requires lots of simulations under different conditions to
know when a configuration is better than another. Due to this fact, parallel
computation is used to reduce the time require to find an approximation to the
global optimal mechanism. So, SA returns the best configuration of a mechanism,
selected among a huge amount of randomly generated configurations, having in
mind that this random process samples intensively the most promising regions
of the search space.
Another optimization method is the Genetic Algorithm (GA) [2]. GA imi-
tates the evolution of the living beings, described by Charles Darwin, and it is
being part of the group of Evolutionary Algorithms (EA). GA are directly, par-
allel, stochastic method for global search and optimization. GA [3] uses genetic
operations such as selection, crossover and mutation to generate new individuals.
Such individuals can be the configurations for the robotic manipulator. As the
species’ evolution in nature, GA tries the development of new and better across
the generations. Since this method is heuristic and not have a defined behavior
and according to the specific problem the number of generations needed to dis-
cover a solution that solves the problem can be lots, the development in parallel
could have an interesting performance comparing it to a sequential GA.
These both methods have been developed previously by several researchers,
e.g. Paradalos et al. [4] describe the implementation of these methods sequential
and parallel. Parallel GA and parallel SA have had better performance mostly
in the execution time comparing to sequential versions working in optimization
problems of several fields (e.g. Traveling Salemen Problem [5], Combinatorial
Optimization [4], Shape optimization [6]).
The method proposed in this article can be used as design tool, although,
it can not consider some features such as manufacturability, standardization,
etc., it intends to approximate the optimum configuration regarding energy con-
sumption subject to a bounded error, then a designer can use this preliminary
configuration in an initial stage of the design process.

1.1 4-Bar Manipulator

The mechanism has four bars and two active joints which angles are denoted
by ϕ1 and ϕ2 , as it is shown in Fig. 2. A parallelogram is formed with the four
elements, one of them is larger in order to be the end effector, which must
follow a given trajectory. The position of the end effector is defined by the direct
kinematics in Eq. (2), the literals L1 and L4 as well as ϕ1 and ϕ2 are those shown
in Fig. 2.
 Comparison of Parallel Versions of SA and GA 293

   
x L1 cos ϕ1 − L4 cos ϕ2
= (2)
y L1 sin ϕ1 − L4 sin ϕ2

Fig. 2. Four-bar manipulator.

The direct kinematics is used to find the position of the end effector given
the angles ϕ1 and ϕ2 , while the inverse kinematics is used to compute the angles
given the end effector position, it is given by Eqs. (3) and (5). Both of them are
used to numerically simulate the manipulator movement to following a given
trajectory.
y L4 φ
ϕ1 = arctan( ) − arctan( ) (3)
x L1 + L4 φ
√
− 1 − d2
φ = arctan( ) (4)
d

ϕ2 = ϕ1 + (φ + π) (5)
x +y −
2 2
− L21 L24
d = cosφ = (6)
1L1 L4
The manipulator velocity is computed via the differential kinematics, shown
in Eqs. (7) to (10).

ẋ = J · q̇ (7)
ẋ = [ẋẏ]T (8)
T
q̇ = [ϕ˙1 ϕ˙2 ] (9)
 
−L1 sin ϕ1 L4 sin ϕ2
J= (10)
L1 cos ϕ1 −L4 cos ϕ2
294 A. H. Baena et al.

Table 1. Inertial data used in the manipulator numerical simulation.

Masses
m1 = 2L1 /0.3 m3 = 0.5L2 /0.15
m3 = 2L3 /0.3 m4 = 2L4 /0.4
Mass centers
lc1 = L1 /2 lc3 = L3 /2
lc3 = L3 /2 lc4 = L4 /2
Moments of inertia
I1 = 0.0226 I2 = 0.00641
I3 = 0.0226 I4 = 0.0226
Proportional control
ui (t) = Kpi · (ϕid (t) − ϕi (t))
Proportional-Derivative control
ui (t) = Kpi · (ϕid (t) − ϕi (t)) + Kpi · (ϕ˙id (t) − ϕ̇i (t))

Finally, the dynamic model considers inertial forces and torques, hence, it is
necessary to calculate the effects of the gravity, masses, lengths, mass centers,
and inertial moments, which are shown in Table 1.
The physics of the mechanism is governed by the Ordinary Differential Equa-
tions (11).

d11 ϕ¨1 + φ1 (ϕ1 ) = τ1

(11)
d22 ϕ¨2 + φ2 2(ϕ2 ) = τ2
Where the functions φi are shown in Eq. (12)

φ1 (ϕ1 ) = g cos ϕ(m1 lc1 + m3 lc3 + m4 L1 )

(12)
φ2 (ϕ2 ) = g cos ϕ(m2 lc2 + m4 lc4 + m3 L2 )
Where dii are constants related with the masses of the elements. They are
defined as shown in Eq. (13).
2 2
d11 = m1 lc1 + m3 lc3 + m4 L21 + I1 + I3
(13)
d22 = m2 lc2 + m4 lc4 + m3 L22 + I2 + I4
2 2

The proportional-integral-derivative control is used to compute the torque

τi , associated with each active join.

τi (t) = kp ei (t) + ki ei (t)dt + kd e˙i (t), ei (t) = ϕid − ϕi (14)
 Comparison of Parallel Versions of SA and GA 295

– The integral and derivatives are solved with numerical methods, in this case,
the Euler method for differential equations for an interval T .
– The integrals and derivatives are calculated in each step t, with the value
obtained of the instant torque computed in the previous step.
– The direct and inverse kinematics are solve each step according to the new
positions and the required intermediate steps in the method.

Finally, we propose a trajectory to test the method. In this experiment we

used a circular path:

X = δx + R cos(ωt)
Y = δy + R sin(ωt)
R = 0.05
(15)
δx = 0.4
δy = 0.3
ω=1

2 Simulated Annealing (SA)

The simulated annealing is a method for approximating the global minimum of
an objective function over a search space, usually in presence of non-linearities
and discontinuities. The SA algorithm is based on the analogy between the sim-
ulation of the annealing of solids and the problem of solving large combinatorial
optimization problems [7].
We use an SA inspired on the Coranna’s version [1] which is shown in
Algorithm 1. This algorithm requires two inputs: initial temperature and can-
didate solutions which are randomly generated in an internal loop. The initial
temperature T0 starts with a great value and it is reduced by multiplying by
a constant coefficient 0 < C < 1 (line 10) to decrease the temperature until it
reach a minimum target value.
The evaluation step is the most computationally expensive in the SA, in
consequence, the parallelization efforts must be focused in this step. The general
workflow of the parallel algorithm works as follows: The sequential SA uses one
thread, given an initial configuration X the SA perturbs N times this config-
uration during the optimizing process. In the sequential algorithm the N per-
turbations are made and evaluated by a single core, in this parallel proposal
we use all cores of the microprocessor. This is the unique modification to the
sequential algorithm, nevertheless, the implications are deeper, considering that
the sequential SA works as follows: it perturbs the initial configuration in one
variable, then, the perturbed configuration is evaluated and compared with the
initial configuration, using the metropolis criterion the perturbed configuration
could be accepted or rejected for replacing the initial configuration, the new ini-
tial configuration is going to be perturbed in the next iteration of this loop and
so on. Notice, that in the sequential algorithm there is an stochastic dependency
between the current perturbed configuration and the next one, on the other
hand, in the parallel version this dependency is removed.
296 A. H. Baena et al.

Algorithm 1: Simulated annealing

Input : T0 , Points
Output: XB
1 X ← RAND(lim inf , lim sup);
2 f(X) ← EVALUATION(X);
3 f(XB) ← f (X);
4 XB ← X;
5 while STOP CRITERION do
6 for temp ← T0 to temp < 0.005 do
7 for t ← 1 to P oints do
8 PARALLEL EVALUATIONS
9 end
10 temp ← temp ∗ C
11 end
12 XB ← X
13 f (XB) ← f (X)
14 end

Our SA algorithm is shown in Algorithm 1, it is similar to the Corona’s

Algorithm but with the parallel evaluation step (line 8).
Given the fact that each evaluation of the perturbed solutions is independent
of the rest, each thread can carry out the evaluation of some of these solutions,
such as it is shown in Algorithm 2. First, temporary variables of X, X̂, f (X)
and f (XB) are declared, as well as an array of disturbed solutions X̂ and its
function f (X̂) (lines 1 to 6). From here, each thread works with its memory
and they just compare (line 21) the current best solution to the best global one,
that is, threads don’t have access to shared memory and, this way, we expect
to reduce lost time doing access to memory in common. When the evaluations
of perturbed solutions end (lines 7 to 20) where all changes and improvements
are stored in temporary variables, the best solution of each thread are compared
with the best global solution and it is selected the best among of all them.

2.1 Stop Criterion

Stop criterion (line 5) is flexible. It is recommended to repeat the SA a few times
until the value of the objective function converges to, approximately, the same
value, this value is stored in f (XB) and is associated with the solution XB.

3 Genetic Algorithm (GA)

Genetic Algorithms (GAs) [8] are optimization techniques based on the Dar-
winian principle of evolution through genetic selection. These kinds of algorithms
have been successfully applied to an ample variety of real-world problems of dif-
ferent complexity. A GA uses an abstraction of the natural selection, using a
 Comparison of Parallel Versions of SA and GA 297

Algorithm 2: PARALLEL EVALUATIONS

input : X, XB, f (X), f (XB), N
output: XB, f (XB)
1 X̂ ← Xtmp ← X
2 XBtmp ← XB
3 f (X̂) ← 0
4 ftmp (XB) ← f (XB)
5 ftmp (X) ← f (X)
6 for i ← 1 to N do
7 ei ← RAND(lim infi , lim supi )
8 X̂ ← X̂ + [0, 0, ..., ei , ..., 0, 0]
9 f (X̂) ← EVALUATION(X̂)
10 if f (X̂) < ftmp (XB) then
11 XBtmp ← X̂
12 ftmp (XB) ← f (X̂)
13 end
/* ACCEPTANCE OF METROPOLIS CRITERIA */
14 Δf ← f (X̂G) − f (Xtmp )
−Δf
15 if Δf ≤ 0 OR {Δf ≥ 0 AND RAND(0, 1) < exp tmperature
} then
16 Xtmp ← X̂
17 ftmp (X) ← f (X̂)
18 end
19 end
20 if ftmp (XB) < f (XB) then
21 XB ← XBtmp
22 f (XB) ← ftmp (XB)
23 end

population of artificial chromosomes. Chromosomes are usually chains of data

that provide information (a real number) about how the individual behaves in
a particular habitat (the problem). The chromosome is evaluated in order to
determine how fitted an individual is, with the purpose of preserving or dis-
carding his genes in the following generations. So, chromosomes are solutions of
the particular problem and the fitness is related or equal to its objective func-
tion evaluation. GAs start with a random population (of chromosomes) which
is evaluated to know what individuals are convenient for the next generation.
GA carries out a process of chromosomes recombination with the best current
chromosomes (parents) to produce a successor population (children), then chil-
dren are mutated, evaluated and used to replace the current population, and
298 A. H. Baena et al.

so on. This process is iterated for a finite number of times (generations). Each
iteration, the solution is expected to improve.
For this GA (Algorithm 3), we propose as an individual of the population
a chain of 7 genes, which represent a candidate configuration for the robotic
manipulator. Therefore, the population POP which the GA works is an array
of the mentioned chains. The optimizing begins with an initial POP generated
randomly of possibles solutions and then, the GA carries out the process of
minimization (lines 3 to 11).

Algorithm 3: Genetic algorithm

input : generations, population, genes
1 POP ← RANDOM(lim inf, lim sup);
2 F ← [0, 0, ..., 0] ;
3 for i ← 1 to generations do
4 F← PARALLEL EVALUATIONS;
5 POP← SORT(POP, F);
6 CHILDREN ← SELECTION(POP);
7 CHILDREN ← RECOMBINE(CHILDREN);
8 CHILDREN ← MUTATION(CHILDREN);
9 POP ← NEW GEN(CHILDREN);
10 end

Like the SA, the evaluation step is parallelized. As it is shown in Algorithm 4,

each thread works with POP and F (lines 1 and 2) temporary arrays. After each
thread carries out a part of the all evaluations, the results are inserted in the
global functions array F, this way, each thread has access to shared memory only
one time.

Algorithm 4: Parallel evaluations

1 Parallel for i = 1, 2..|P OP | do
2 Fi = EV ALU AT ION (P OPi )
3 end

4 Results

In this section, the results of our experiment are described. Since our purpose
is comparing parallel versions of the SA and the GA, and as each of both algo-
rithms works with parameters which determine how they carry out the process,
we must use conditions where the algorithms behave of an adequate way. The
conditions which govern the performance of our SA are initial temperature,
amount of points, times the SA works and, of course, the number of cores used.
 Comparison of Parallel Versions of SA and GA 299

‘Times’ is the value to which the stop criterion mentioned in previous sections
is subject. To find out which are the adequate conditions for the experiment,
we carry out preliminary tests with different parameters which are shown in
Table 2. Analyzing the data, It can be noted that the configuration where values
are 125, 50 and 10 (for temperature, points and times, respectively) has the
lowest average for the objective function, as well as the lowest standard devia-
tion. Furthermore, the execution time that the SA takes with these parameters
is below of the arithmetic mean for recorded times. Thus, we opted to take the
aforesaid configuration like the conditions for true tests.

Table 2. Objective function varying the conditions for the SA

Points Temps Times Mean SD Max Min Time (Mean)

25 25 5 6594.514 578.2811 7241.312 6127.422 67.79439
25 25 10 7403.964 919.7598 8450.774 6725.321 135.4295
25 75 5 6118.325 187.3511 6333.709 5993.086 80.83644
25 75 10 5948.943 63.07227 6018.22 5894.847 155.4485
25 125 5 6449.567 473.9563 6865.875 5933.76 80.895
25 125 10 8100.752 1291.439 8890.017 6610.396 690.752
25 175 5 6661.187 678.9761 7390.485 6047.335 90.40371
25 175 10 7655.782 237.4511 7929.832 7511.314 196.5836
50 25 5 596.5836 11.1195 206.8913 184.7998 141.0356
50 25 10 6992.395 1590.787 8816.63 5893.924 265.6407
50 75 5 7140.305 604.4596 7826.031 6684.72 151.7968
50 75 10 6725.831 330.8201 7099.958 6471.954 310.9205
50 125 5 6338.199 504.8088 6883.95 5887.985 156.6621
50 125 10 5903.71 18.63697 5925.194 5891.894 312.5321
50 175 5 6126.922 253.421 6393.968 5889.779 162.3664
50 175 10 6290.447 411.5418 6704.64 5881.609 330.7269
75 25 5 6167.833 359.2555 6571.708 5883.871 229.0864
75 25 10 6023.295 116.8177 6113.512 5891.342 392.8536
75 75 5 6368.638 490.9635 6893.742 5921.034 224.4166
75 75 10 6523.281 855.9222 7509.494 5974.127 459.3084
75 125 5 6485.036 185.7312 6612.42 6271.925 240.5058
75 125 10 6853.661 536.2561 7472.461 6524.623 480.7757
75 175 5 6122.928 363.733 6542.868 5906.702 239.6924
75 175 10 6447.443 3359.526 6845.913 607.7692 500.4325

Moreover, for our GA, the conditions which govern its performance are the
amount of population and generations, as well as the number of cores used.
For the GA, tests were executed with different combinations of generations
and population which are shown in Table 5. Like the SA tests (Table 2), we
considerate as the better parameters for the true tests, the conditions where the
300 A. H. Baena et al.

GA has the lowest average, lowest standard deviation and where the average
time is below of the arithmetic mean. The configuration, that describes the
previous description, is the GA working with 300 generations and a population
of 175 individuals. Even though the configuration previously described fulfills
the necessary criteria, we consider more feasible to use 75 individual for the
population through 400 generations given that it has almost the same statistics
and by considerably less time.

Table 3. Objective function statistics for the SA

Threads Points Temp Times Mean SD Max Min

8 50 125 10 5968.21375 106.0896113 6224.180244 5899.072774
1 50 125 10 5975.654339 128.6481023 6301.167460 5898.078654

Table 4. Execution time statistics for the SA

Threads Points Temp Times Mean SD Max Min

8 50 125 10 312.4205 7.325753659 319.9205 302.9205
1 50 125 10 775.2191731 75.24748331 807.231196 698.236

The results of the true simulations were summarized in statistics concerning

the minimization of the objective function and the execution time. These statis-
tics are shown in Tables 3 and 4 for the SA and they are shown in Tables 6 and
7 for the GA.
Additionally, in Fig. 3 (For the SA) and Fig. 4 (For the GA) can be shown how
the objective function behaves through the minimize process for both algorithms.
These graphs show how GA tends slightly faster to a global minimum, while SA
tends to have a less steep slope. Something equally interesting is how the SA
behaves when the temperature is very close to zero, how in these temperatures
the minimum decreases a little more, something that in higher temperatures it
carries out in a small scale.
 Comparison of Parallel Versions of SA and GA 301

Table 5. Objective function varying the conditions for the GA

Gens Population Mean SD Max Min Time (Mean)

100 25 5948.81 87.4562956 6123.066 5896.562 3.064973
100 75 5896.105 11.6481354 5916.619 5882.258 8.988048
100 125 5887.878 4.318659675 5896.305 5883.955 14.15521
100 175 5886.633 3.91596454 5893.639 5883.16 19.70656
100 225 5884.212 1.550943486 5886.635 5882.294 26.12352
200 25 5940.722 50.01127077 6030.216 5892.125 5.847065
200 75 5884.957 1.111173328 5886.36 5883.617 16.99983
200 125 5886.275 4.328700825 5892.569 5882.312 28.5828
200 175 5884.772 2.843259643 5889.056 5881.904 39.30059
200 225 5885.181 3.090062037 5890.908 5882.93 51.91625
300 25 5913.216 14.55270843 5933.787 5897.383 8.731338
300 75 5885.806 5.310659254 5895.842 5881.749 25.53023
300 125 5883.863 2.416952464 5888.347 5881.573 42.11146
300 175 5881.653 0.360897382 5882.147 5881.161 58.56627
300 225 5882.314 0.881892134 5883.541 5881.429 79.4524
400 25 5904.799 18.5366548 5927.339 5884.802 11.71828
400 75 5882.036 0.98095959 5883.272 5880.977 34.01984
400 125 5882.242 1.057190353 5883.391 5880.835 56.1136
400 175 5882.643 0.612921241 5883.162 5881.497 78.35479
400 225 5881.919 1.237379279 5883.766 5880.778 104.978
500 25 5917.184 40.0366268 5992.548 5887.238 14.46798
500 75 5883.117 1.694247849 5885.302 5881.585 42.54183
500 125 5883.581 2.075564146 5886.204 5880.886 69.93083
500 175 5882.511 1.804717969 5885.561 5880.928 97.76304
500 225 5881.592 0.521225604 5882.31 5880.967 134.5352

Table 6. Objective statistics for the GA

Threads Gens Population Mean SD Max Min

8 400 75 5881.103 2.19020944 5888.697 5880.895
1 400 75 5882.243 1.315935506 5881.456 5880.88

Table 7. Execution time statistics for the GA

Threads Gens Population Mean SD Max Min

8 400 75 34.37404 0.398747021 35.54596 33.65045
1 400 75 155.5839 0.883990483 156.2735 154.5873
302 A. H. Baena et al.

Fig. 3. Objective function behavior with the GA.

Fig. 4. Objective function behavior with the SA.

5 Conclusions

Firstly, the computation time, as expected, is lower when various threads carry
out the optimization process in contrast with a single one. Nevertheless, there
is a significant variance in the computation times among different executions
 Comparison of Parallel Versions of SA and GA 303

with the same parallelization, in all cases the parallel version shows a significant
time reduction. This is due to the adequate implementation and selection of the
parallel strategy. We parallelized the most time-consuming task, additionally, we
take care of using a low amount of memory, by reusing variables and avoiding
to preserve results that are no longer needed inside the loops with the purpose
of maximizing the cache use and minimizing the request of memory updating to
the RAM.
The analysis of the data from the previous section provides enough informa-
tion to assert that GA is a more convenient algorithm for the optimization of
the robotic manipulator. Comparing the SA with the GA, the second converges
faster to the best solution. Besides, the total processing time of the GA to is
significantly less than that used by the SA. Another important consideration
about the comparison is that the SA solutions present a huge variance among
executions, while the GA reports more similar objective function values of the
best solutions from different executions, that indicates that the GA is a more
confident algorithm. Future work considers the comparison of other optimization
algorithms for this problem.

References
1. Martini C, Ridella S, Corana A, Marchesi M (1987) Minimizing multimodal func-
tions of continuous variables with the “simulated annealing” algorithm. ACM Trans
Math Softw 13:262–280
2. Popov A (2005) Genetic algorithms for optimization. User Manual, Hamburg 2013
3. Chen Z, Zhang Z, Xie J, Guo Q, Tao Y (2019) Metaheuristic optimization method
for compact reactor radiation shielding design based on genetic algorithm. Ann Nucl
Energy 134:318–329
4. Pardalos PM, Pitsoulis L, Mavridou T, Resende MGC (1995) Parallel search for
combinatorial optimization: genetic algorithms, simulated annealing, tabu search
and GRASP. In: Parallel algorithms for irregularly structured problems, pp 317–
331. Springer, Heidelberg
5. Suh JY, Van Gucht D, Prasanna J (1991) Parallel genetic algorithms applied to the
traveling salesman problem. SIAM J Optim 1:515–529
6. Brauer H, Ziolkowski M (2005) Shape optimization with adaptive simulated anneal-
ing and genetic algorithms. In: Wiak S, Krawczyk A, Trlep M (eds) Computer engi-
neering in applied electromagnetism. Springer, Dordrecht, pp 25–30
7. van Laarhoven PJ, Aarts EH (1987) Simulated Annealing: Theory and Applications.
Springer, Netherlands
8. McCall J (2005) Genetic algorithms for modelling and optimisation. J Comput Appl
Math 184:205–222
 Dynamic Modeling and Control of a Pivoting
Nozzle

Christopher Virgen-Vázquez(B) , Eusebio Eduardo Hernández,

and Armando Oropeza Osornio

ESIME Ticomán, Instituto Politécnico Nacional, Ciudad de México, Mexico

tablaton@gmail.com

Abstract. This proposal is based on the dynamic analysis of the orientation mech-
anism of a pivoting nozzle, that is composed of a bell-type nozzle and three actua-
tors mounted and positioned at 120° respectively, for the orientation of the output
thrust of a propelled system. The behavior of the elements together will be studied
to obtain the dynamic model, generating the equations that describe the move-
ment of the actuators, and thus evaluate a preliminary law of control that allows
the correct interaction of the entire system.

Keywords: Propulsion · Vector · Nonlinear system · Law of control · Nozzle ·

Pivotal · Actuator stabilization

1 Introduction

Originally the idea of designing a control law for a pivoting nozzle is the need to be able
to implement orientation control in propelled systems, which allows us to orient the exit
thrust. Recently, in thrust nozzle vectoring area, there was a proposal made by Naguil,
Jorge L. Pedroni, et al. [1], which sets out the dynamic considerations on the control of
a pivoting nozzle, however, in this case they analyzed a system with only two actuators
that are positioned perpendicularly. Other references have mentioned the aerodynamics
and evaluation of performance of this kind of systems, Schaefermeyer [2] and Berrier
and Taylor [4], for rockets mainly, and the simulation of the mass flow control of a
common system, Ruoyu Deng, Fanshi Kong and Heuy Dong Kim [3].
For the development of the proposal, initially it was necessary to obtain the dynamic
equations of the pivoting nozzle system. Subsequently, we proposed and evaluated a
control scheme based on the dynamic model. Lagrange mathematical modeling has
been used to obtain dynamic equations. For the control law it was expected to obtain
a non-linear model and use the approximate linearization method and thus propose a
control law.
At the beginning of the mechanical design, three symmetrically located actuators
were used, which improve the movement in the nozzle space and address the outflow
(Fig. 1). The system is composed of a bell-type nozzle and three actuators mounted and
positioned at 120° respectively, for the orientation of the output thrust of a propelled

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 304–309, 2020.
https://doi.org/10.1007/978-3-030-45402-9_29
 Dynamic Modeling and Control of a Pivoting Nozzle 305

Fig. 1. Proposed model.

system. The deeper analysis of oscillations and vibrations will also be left for the future,
with the corresponding control of these phenomena.
Therefore, a mechanical design and its control law are proposed, which allow to
address the flow of the nozzle, considering the dynamic and kinematic constraints.

2 Governing Equations
Considering the mechanism of Fig. 1, we should analyze an over-actuated mechanical
system, which has external forces acting on it. We must find the mathematical model
that describes the behavior of the system. After, it is necessary to design control for the
mechanism that will help to maintain the system variables in their nominal operating
values, for existing or in the presence of disturbances. The Lagrange method has been
used to obtain a dynamic representation of the mechanism. In Fig. 2, reference and
rotated frames are presented.

Fig. 2. Origin point, angles and related axes.

306 C. Virgen-Vázquez et al.

Kinetic Energy. The equation of kinetic energy for the system can be written as next
equation [5]:
1  2 
T = I x x θ̇1 + θ̇22 (1)
2
where:
I x x represents the moment of inertia coefficient.
θ1 is the first angle to rotate.
θ2 is the second angle to rotate.
Potential Energy. Potential energy only involves the force of gravity, mass and
displacement at two angles, and can be defined by next equation:

V = −mgl cos θ1 cos θ2 (2)

where:
m represents the mass.
g is the gravity constant.
l is the distance from the origin to the point of analysis.
Lagrangian Equations. In the next, we formulate Lagrange’s dynamic equations for
the system, where Q = 0. By defining q1 = θ1 , q2 = θ2 , we obtain next equations.

I x x q̈1 + mgl sin q1 cos q2 = Q 1

I x x q̈2 + mgl cos q1 sin q2 = Q 2 (3)

Generalized Forces. The complete equations of the generalized forces are given by a
set of equations as follows:
 
I x x q̈1 + mgl sin q1 cos q2 − f 1 l x y sin q2 − l cos q2 sin q1
  √
l l 3
+ f 2 − sin q2 sin q2 x2y + l cos q2 + cos q1 x y2
  √
l l 3
+ f 3 − sin q1 sin q2 x2y + l cos q2 + cos q1 x y2
  
I x x q̈2 + mgl cos q1 sin q2 − f 1 − cos q1 l x y cos q2 − l sin q2
  
lx y
+ f2 2 cos q2 − l sin q2 cos q1
(4)
  
lx y
+ f3 2 cos q2 − l sin q2 cos q1
 Dynamic Modeling and Control of a Pivoting Nozzle 307

3 Control
The problem of controlling a pivoting nozzle is presented, where the behavior of a non-
linear model of the system around a given operating point is analyzed (in our case, where
the corresponding angles are zero, which is the vertical position).
In the case of a highly non-linear system, with more outputs that input variables, the
modeling in the state space could be suitable [6, 7]. This implementation can be obtained
as next equations.

x 1 = q1 ẋ1 = x2
− mgl
Ix x sin x1 cos x3 + − l cos x3 ) sin x1 ) +
I x x u 1 ((sin x 3
1

 √ 
1 lx y lx y 3
x2 = q̇1 ẋ2 = u 2 − sin x3 sin x3 + l cos x3 + cos x1
Ix x 2 2
 √ 
1 lx y lx y 3
+ u 3 − sin x1 sin x3 + l cos x3 + cos x1
Ix x 2 2
x 3 = q2 ẋ3 = ẋ4
mgl 1   
− cos x1 sin x3 + u 1 − cos x1 l x y cos x3 + l sin x3 +
Ix x Ix x
x4 = q̇2 ẋ4
mgl 1   
=− cos x1 sin x3 + u 1 − cos x1 l x y cos x3 + l sin x3
Ix x Ix x
 
1 lx y
+ u2 cos x3 − l sin x3 cos x1
Ix x 2
 
1 lx y
+ u3 cos x3 − l sin x3 cos x1 (5)
Ix x 2

Where, the displacement angles θ1 and θ2 or also called q1 and q2 , are defined as state
variables, as well as their respective angular velocities. The input variables ui represents
the force that each actuator exerts on the ring, in order to move it and control the plane
as is shown in Fig. 1.
Considering the model of Eq. (5) and taking into account that all the state variables
are measurable, we can use them for feedback control. We define the disturbed variables
around the nominal equilibrium values. For the initial evaluation of the proposed control
scheme of the pivoting nozzle, the approximated linearization procedure around the
equilibrium point are presented in Eqs. (6) and (7), this procedure is based in references
[9–11].
⎡ ⎤
01 0 0
∂f ⎢ − mgl 0 0 0 ⎥
⎢ ⎥
A= |xi ,u i =(0,0) = ⎢ Ix x ⎥ (6)
∂x ⎣ 0 0 0 1⎦
mgl
0 0 Ix x 0
308 C. Virgen-Vázquez et al.

⎡ ⎤
0 0 0
⎢ l √ l √ ⎥
∂f ⎢ x y 23 x y 23 ⎥
⎢ l
− ⎥
B= |xi ,u i =(0,0) = ⎢ Ix x Ix x Ix x ⎥ (7)
∂u ⎢ 0 0 0 ⎥
⎣ lx y lx y ⎦
l
− Ixxyx 2 2
Ix x Ix x

Physically, this means that we can take the nozzle to the horizontal position, as long
as we bring the nozzle to the ideal equilibrium position, in that case zero for both angles,
which is its ideal operating position. We propose and evaluate a control law for trajectory
tracking [6, 8], with y1 = x(1), y2 = x(3), yd = sin(t) and ẏd = cos(t), and signals of
control in the form u 1,2,3 = (K 1 ∗ ex ) + (K 2 ∗ e y ), where ex and e y are signals of error
between state variables and trajectory variables.

4 Results

The system was compared in three different scenarios, the open loop system (nonlin-
ear), closed loop system (with the proposed control scheme) and the linearized system.
Initially, the simulation of the open loop system was carried out, without considering
external forces affecting the system. Subsequently, the simulation of the linearized sys-
tem was performed, and the response of the system was compared according to the
definition, characteristics and advantages of the approximate linearization. Finally, the
simulation of the controlled system was carried out, where one can see the advantages
of using the proposed control law (Fig. 3).

Fig. 3. Stabilized system with 10-degree disturbance.

 Dynamic Modeling and Control of a Pivoting Nozzle 309

5 Conclusions
Once the mathematical model by using Lagrange method has been obtained, an analysis
of the generalized forces was required, this process was somewhat complex because
the forces acting were determined and it was analyzed how these forces interact with
the system. Thus, the mathematical model was completed and finally it was modeled
in the form of state variables, a simple process that requires a complete and conscious
prior analysis for the system. Then, we approached a pivoting nozzle that represents a
non-linear model and its behavior at its point of operation, that for our use is around zero.
Since the model is highly non-linear, an approximate linearization method was presented,
which is useful near the point of equilibrium and low disturbances. Finally, a control
scheme was evaluated for trajectory tracking, where rapid stabilization was obtained.
This result was compared with them from the open loop and closed loop system, as well
as the linearized system, and it was showed that the control law stabilizes the system in
a faster and more efficient way.

References
1. Naguil JL, Pedroni JP, Cova, WJD, Consideraciones dinámicas sobre el control de una tobera
pivotante
2. Schaefermeyer, MR (2011) Aerodynamic thrust vectoring for attitude control of a vertically
thrusting jet engine. Utah State University Logan, Utah
3. Deng R, Kong F, Kim HD, Numerical simulation of fluidic thrust vectoring in an axisymmetric
supersonic nozzle. Department of Mechanical Engineering, Andong National University,
Andong, Korea, pp 760–749
4. Berrier BL, Taylor JG (1990) internal performance of two nozzles utilizing gimbal concepts
for thrust vectoring, NASA TP-2991
5. Taylor JR, Díaz JID, Grau AC (2013) Mecánica clásica. Reverté, ISBN 9788429143126
6. Slotine J-JE, Li W, et al (1991) Applied nonlinear control, vol 199, Prentice Hall Englewood
Cliffs, NJ
7. Khalil HK (1996) Nonlinear Systems. Prentice-Hall, New Jersey, vol 2(5), pp 5–1
8. Ogata K, Yang Y (1970) Modern control engineering. Prentice-Hall Englewood Cliffs, NJ
9. Sira-Ramirez H, Agrawal SK (2004) Differentially flat systems. CRC Press
10. Sira-Ramírez, H, Márquez, R, Rivas-Echevarría, F, Llanes-Santiago R (2005) Control de
sistemas no lineales: linealización aproximada, extendida, exacta. Pearson Prentice Hall
11. Cortés-Romero J, Sira-Ramirez, H, Luviano-Juárez A (2011) Control lineal robusto de sis-
temas no lineales diferencialmente planos. RIAI Revista Iberoamericana de Automatica e
Informatica Industrial
 A Closed Loop Controller Design Which Can
Minimize the Wind Effects on a Parabolic Solar
Tracker

J. Díaz-Salgado1 , O. M. Huerta-Chávez2(B) , E. A. Villegas Jimenez1 ,

E. Enciso-Contreras1 , L. A. Cortés-Marín1 , and C. Ortiz-Cil3
1 Tecnológico de Estudios Superiores de Ecatepec, Edo de Méx Ecatepec de Morelos, Mexico
jorgesalgado@tese.edu.mx, llanus_2@yahoo.com,
eenciso@hotmail.com, lualcoma1@gmail.com
2 Cátedras CONACyT, Tecnológico de Estudios Superiores de Ecatepec,
Edo de Méx Ecatepec de Morelos, Mexico
lecram@hotmail.com
3 Tecnológico de Apan Hidalgo, Apan, Mexico

ceoc_ceoc@hotmail.com

Abstract. In the present work, it was conducted a steady-state fluid analysis

to obtain the aerodynamic coefficients of a Parabolic Solar Tracker (PST) for
different azimuth and elevation angles. A computational fluid dynamic (CFD)
wind flow analysis technique was applied to model an extreme weather condition
called North Wind (NW), which can severely affect the PST under operational
conditions. The North Wind case simulated was validated using experimental data
without bottom effects in order to verify the degree of approximation between the
results obtained with CFD approach. The obtained aerodynamic coefficients were
used to: (i) establish both analytical and numerical models for each PST degree
of freedom for dynamic analysis and simulation purposes, (ii) design different
closed loop controllers which can minimize the wind effects and (iii) to establish
a correct dimensioning of the control actuators.

Keywords: Parabolic Solar Tracker · CFD · Solar energy and control

1 Introduction
Solar energy is one the most attractive renewable energy because is unlimited, clean and
free and its development permits a sustainable growing of alternate countries. Among
the different forms of high-temperature solar energy harnessing Disc-Stirling systems
present the highest solar-electric conversion efficiency thus having the potential of com-
ing through the cheaper sun energy systems [1]. In a Disc-Stirling system the Parabolic
Solar Tracker (PST) is the subsystem tasked to concentrate the solar radiation energy into
a receiver that generally transfers such energy to a heat engine [2]. To achieve the high
temperature a PST must have a two degree of freedom sun tracking system (azimuth,
elevation) along the dairy operation.
© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 310–326, 2020.
https://doi.org/10.1007/978-3-030-45402-9_30
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 311

The presence of extreme weather conditions in areas with important solar energy
resource is common. As PST are usually installed in open areas they suffer commonly
this extreme wind conditions that might affect the efficiency and/or the integrity of the
PST. Solar power companies have recognized that the wind is a fundamental element
in the operation of a PST in three different forms: (i) the efficiency reduction due to
the misalignment of the desired trajectory caused by forces due to the wind loads, (ii)
temperature losses caused by the convection phenomenon due to the wind speed and (iii)
the severe damage over the PST and/or its mechanical structure as a result of a critical
wind condition [3–5].
In this work, the effect over a PST of an extreme wind condition named North
Wind (NW) that is originated from a wind blowing from the north to the south, very
common on winter in the mexican coast that surrounds the Gulf of Mexico, reaching
wind speeds up to 22 m/s is studied [6]. A steady-state computational fluid dynamic
(CFD) wind flow analysis technique is applied to obtain the aerodynamic coefficients
of a PST for different azimuth and elevation angles along a test day with the NW
condition. Experimental data were used to validate the simulated results obtained with
CFD approach. Finally, the obtained aerodynamic coefficients were used to: (i) establish
both analytical and numerical models for dynamic analysis and simulation purposes, (ii)
to design a closed loop controller which can minimize the wind effects, and (iii) establish
a correct dimensioning of the control actuators.
As the azimuth and elevation controllers were designed in a decoupled form, three
different controllers were designed and compared to select the most appropriated for
each case. The proposed controllers were designed in terms of the physical parameters
of the PST. This feature permits to apply the same control law to another PST with
different dimensions but with the same geometrical form.
This work is organized as follows: first an adequate numerical and analytical model
is proposed for the PST, and its unknown parameters are identified; next the aerodynamic
momentum coefficients are determined by the application of a (CFD) wind flow analysis
technique under a NW weather condition; in the next section are designed different sun
tracking controllers for the PST, and compared in nominal and perturbed conditions
simulations in the subsequent section; finally some conclusions are drawn.

2 Parabolic Solar Tracker Modeling

In this section the mathematical and numerical models of the PST are established. The
physical parameters (Table 1) of the PST considered (Fig. 1) makes it suitable for medium
temperature applications [7].
312 J. Díaz-Salgado et al.

Fig. 1. Parabolic Solar Tracker

The PST was designed and manufactured with aluminum alloy 1060 and has two
DC motors that generates two torques (τ1 , τ2 ) to control the azimuth θ angle of the base
and the elevation φ angle of the parabolic plate, respectively.

Table 1. Physical parameters of the PST considered.

Parameter Value Symbol Units

Reflecting area 10.10 AT m2
Focal distance 1.67 lf m
Projected area 4.59 AP m2
Total mass 2.32 × 103 m T kg
Focal to diameter ratio 0.47 FD
Concentrating ratio 10.00 η

For simplicity two decoupled models are proposed one for the base and other for the
parabolic plate of the PST.

2.1 Mathematical Model

Base Model
The rotation of the circular PST base produces the azimuth angle γ . Since in the consid-
ered PST the base is three times heavier that the parabolic dish and in closed loop it is
expected a slow motion of the dish with respect to the base this second one is modelled
as an over-damped rotational inertia system with a linear second order space state model
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 313

(1) that arises when a DC motor is coupled with a rotational inertia load.

ẋ1 = x2 (1a)

ẋ2 = − px1 + kτ1 (1b)

yγ = x1 (1c)

x1 = γ , x2 = γ̇ , u 1 = τ1 (1d)

where k is the input gain, and p is the not dominant pole.

Dish Model
The rotational movement of the parabolic dish produces the elevation angle θ . Because
of the symmetry of the dish this is considered as a pendulum with a defined center of
mass and so a non-linear second order model in space state is used as the mathematical
model of the dish (2).

ẋ3 = x4 (2a)

ẋ4 = −a1 x4 − a2 sin(x3 ) + a3 u 2 (2b)

yφθ = x3 (2c)

x3 = θ, x4 = θ̇, u 2 = τ2 (2d)

where a1 , a2 , a3 can be expressed in terms of physical parameters γ

b g 1
a1 = 2
, a2 = , a3 =
ml f lf ml 2f

where m is the dish mass, l f is the parabolic dish focal length, b is the viscous damping
coefficient given as the sum of the bearings, gears and motor viscous friction, and g is
the gravitational constant.

2.2 Numerical Model

The numerical model of the PST is obtained exploiting the fact that many CAD special-
ized mechanical design software can formulate and solve the movement equations of a
mechanical system only from the geometry and dimensions of the physical elements and
the type of material it is composed. In particular the methodological design followed was:
(i) the drawing, assembly, and configuration of the materials of the mechanical elements
of the system in a CAD software, (ii) the use of a simulation mechanics toolkit to export
the mechanical model, including the joints, body elements, frames, and transforms to a
numerical simulation software, and (iii) the inclusion of the gravitational forces in the
numerical model. Different dynamic open and closed loop response simulations can be
done with this numerical model and its outputs are considered as the actual outputs of
the system.
314 J. Díaz-Salgado et al.

2.3 Parameter Identification

The parameters (k, p) of the base and (a1 , a2 , a3 ) of the dish are identified by using the
numerical model outputs as the measured signals of a recursive least squares parameter
identification method. This can be done when the mathematical model can be expressed
as a linear regressor (3) in the unknown parameters:

y(k) = Ψ T (k)Θ (3)

The measured discrete time outputs generated by the numerical model ẏγ (k), yθ (k)
and ẏθ (k) are used to express (1b) and (2b) as linear regressors (4) and (5):
 
  p
ÿγ (k) = − ẏγ (k) u 1 (k) = Ψ1 (k)Θ1 (4)
k
⎡ ⎤
  a1
ÿθ (k) = − ẏθ (k) −sin[yθ (k)] u 2 (k) ⎣ a2 ⎦ = Ψ2 (k)Θ2 (5)
a3

Where y(k) is the measured signal vector of the system, Ψ (k) is the regression matrix
constituted by known function observations, and Θ is the unknown parameter vector.
The variable k denotes the discrete time index.
The scalar least square method has the next form [8]:
P(k − 1)Ψ (k)e(k)
Θ̂(k) = Θ̂(k − 1) + (6)
1 + Ψ (k − 1)T P(k − 1)Ψ (k)
P(k − 1)Ψ (k)Ψ (k)T P(k − 1)
P(k) = P(k − 1) − (7)
1 + Ψ (k − 1)T P(k − 1)Ψ (k)
e(k) = y(k) − Ψ (k)T Θ̂(k − 1) (8)

Where P(k) is a positive definite covariance matrix, e(k) is the prediction error,
and Θ̂(k) is the estimated parameter vector. The followed methodology considers in
applying two different tests: (i) the parabolic dish free response from the upper position
and zero velocity as the initial condition, and (ii) the unit step response of the base from
zero position and velocity initial condition.
The simulated free response (Fig. 2) for the parabolic dish (γ , γ̇ ) and unit step forced
response for the base θ, θ̇ shows good agreement between the proposed numerical and
analytical models, in both transient and steady state responses, and makes the proposed
model (1-2) suitable for control design purposes.
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 315

Fig. 2. Comparison between the numerical and analytical model responses.

3 CFD Wind Flow Analysis

A computational fluid dynamic (CFD) wind flow analysis technique is applied to the
PST CAD model to determine the aerodynamic momentum coefficients under an NW
extreme weather condition [9]. As mentioned above, the CFD Fluent software was used
to perform the fluid flow simulations of the wind loads over the PST. The incompressible
Navier-Stokes equations were solved in a classical model based on Reynolds Average
formulation, better known as RANS. Therefore, momentum and continuity equations
were coupled to solve the velocity field. The Spalart Allmaras (S-A) viscous model was
applied to solve a single partial differential equation (PDE). In the simulations, the S-A
turbulent model was considered which as most turbulence models follow Boussinesq’s
assumption and the Reynolds stresses are considered proportional to the fluid shear
stresses (see Čedomir Kostić 2015).
The S-A viscous model is not typically applied in the literature for PST problems,
because is a special model for aeronautical applications in the aerospace industry, how-
ever, it has been identified that the S-A model works correctly under the following two
scenarios: 1. For a fine enough resolution to completely resolve viscous sub-layer (but
not so fixable for Re~10E6) and 2. Putting the first grid point in the log-region (z+~30-50)
which need high grid intensity near wall (see Gerasimov, A.V., 2003).

3.1 CFD Results

The residual behavior computed from the S-A model 8, where 8( 8 is the convergence
criteria used to stop the solver, e.g. 10_8 → 1x10ˆ8, for the most significant case there
were simulated 18 h. All data are based on absolute value and the same process was
applied in all cases from 6 to 18 h. Also, there were conducted simulations with S-A
model 4 which have not sufficient accuracy in results.
316 J. Díaz-Salgado et al.

To obtain better agreements in the results, the CFD team has been followed
professional recommendations with a new set of simulation with S-A 10.
Plots in Fig. 5 from a to c shows not discrepancy between S-A 8 and S-A 10,
its mean no more than 3% error for each data point. Actually, S-A 10 behavior tend
to RKE 8 results and the differences between plots are not significant in results, both
models have a good quality, quantity and great convergence. However, the computational
cost are more expensive for S-A 10, the number of cells needed increase from 5 to 123
million and the computational time from 6 to 28.4 h with more than 18 000 iterations
each case (working with 256 GB of RAM and 48 processors in a Professional Server
from TESE Institute).

3.2 Experimental Test

In order to validate the simulation results, it was performed laboratory experimental

work. Two different numerical methods were compared with the experimental data which
was obtained with a wind channel (Andre, Mier-torrecilla, & Wüchner, 2015). In this
work only the results of the S-A 8 method were compared with experimental data.
: (i) not significant differences were founded with respect the RKE 8 method and
(ii) the first one is as twice as fast than the second one.The PST-prototype used in this
study was made of ABS-HI-100 plastic material (see Fig. 6). l. The geometric similarity
was scaled by a constant factor CGF 1:4.51 and the Reynold Number (Re~2.54E6) to
scale all forces (e.g., viscous and inertial forces) in the prototype was used as a dynamic
similarity parameter.

3.2.1 Results
Comparisons between the total forces from experimental measurements vs numeri-
cal results were qualitative and quantitative favourable in most cases, with predictions
usually within 77% of the measured data as clearly shown in Fig. 3a–b.
During the analyses, there were identified two particular and visible behaviours over
the curves obtained from a sixth order polynomial approximation. First, Fig. 3a shows
a phase shift between the experimental measurements and the numerical data curves
and second, Fig. 3b, shows more emphasized behavior in the valleys and crests of the
experimental measurement over the curve. Quantitative differences were found at some
data points in the graph due to the effects of significant blockages during the rotation
of the dish within the wind test section, since the area of obstruction of the prototype
within the wind channel varies. However, qualitatively the behavior between numerical
vs analytical results is very similar, which gives a great validity to the numerical results.
In addition, during experiments, three particular altitude and azimuth angle positions
from the PST at 11 h (50º, 135º), 12 h (63º, 140º) and 18 h (45º, 320º) presented oscillation
forces around -3 to 3% of medium effective velocity. In fluids mechanics, the effect is
called vortex shedding, where the vortices are created at the back of the dish and detach
periodically from either side of the dish due alternating low pressure on the downstream
side.
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 317

Fig. 3. a.–b. Represent time vs Fx and Fy wind total loads simulated compared with experimental
total loads over the PST.

4 Control Algorithms Design

Three different control algorithms are designed for each PST angle. Departing from the
static and dynamic performance in closed-loop response one of the three algorithms is
selected for each angle.

4.1 Parabolic Dish Controllers

Since the mathematical model of the dish is non-linear, two of the proposed control algo-
rithms are designed by standard linearization method around the upward dish position,
meanwhile the last is anon linear one.
318 J. Díaz-Salgado et al.

Model Linearization
The upward parabolic dish position corresponds to an elevation angle θ = 0. To obtain
the linearized model the Jacobian matrices A and B are defined as:
 
∂ f1 ∂ f1  ∂ f1 
∂ ∂  ∂u 
A = ∂ f23 ∂ f24  , B = ∂ f12 
x x (9)
∂x ∂x
 ∂u

3 4 θ=0 2 θ=0

Where:

f 1 = x4 (10a)

f 2 = −a1 x4 − a2 sin(x3 ) + a3 u 2 (10b)

yielding the linearized model

   
0 1 0
ẋ = x+ u (11a)
−a2 −a1 a3
 
y= 10 x (11b)

State Feedback + Integral Control Algorithm

The first control designed is a state feedback + integral control algorithm [10]. The
design consists in determining a gain feedback vector K by proposing a control signal
as u = r − K x and forcing a desired dynamic response. The addition of the integral part
permits to eliminate the tracking error simultaneously to a desired dynamic performance.
The first stage consist in the definition of an additional element x N of the space state
vector and an integral gain ki yielding:
      
ẋ ( A − B K ) Bki x 0
= + r (12)
ẋ N −C 0 xN 1
With the substitution of A, B and C gives:
⎡       ⎤
  0 1 0 0    
ẋ − K ki ⎦ x 0
= ⎣ −a2 −a1 a a + r (13)
ẋ N   3 3
xN 1
− 10 0

The new space state matrix An is:

⎡ ⎤
0 1 0
An = ⎣ −a2 − a3 k1 −a1 − a3 k2 a3 ki ⎦ (14)
−1 0 0
With this matrix the proposed closed loop third grade polynomial is stated:
⎡ ⎤ ⎡ ⎤
 s00 0 1 0 

|s I − An | = ⎣ 0 s 0 ⎦ − ⎣ −a2 − a3 k1 −a1 − a3 k2 a3 ki ⎦
 00s −1 0 0 
 
 s −1 0 


=  a2 + a3 k1 s + a3 k2 + a1 −a3 ki  = s 3 + (a3 k2 + a1 )s 2 + (a3 k1 + a2 )s + a3 ki
 1 0 s 
(15)
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 319

A desired closed loop dynamic is proposed

s 3 + 90.06s 2 + 95.48s + 23.85 (16)
Finally the desired and proposed closed-loop dynamics are equated (15) = (16) and
the gains determined
95.48 − a2 90.06 − a1 23.85
k1 = , k2 = , ki = (17)
a3 a3 a3
Optimal Control
This algorithm starts with a similar control signal proposition as in the state feedback
case: u = −K x. But in this case the gain vector is determined by solving a cost function
that minimizes the state trajectory and at the same time the control effort:
∞
J = ∫ x ∗ Qx + u ∗ Ru dt (18)
0
Where Q is a symmetric semi-definite positive matrix and R is a symmetric definite
positive one. The gain vector is determined from
K = R −1 B ∗ P (19)
Where P is a symmetric definite positive matrix calculated by solving:
(A − B K )∗ P + P( A − B K ) = − Q + K ∗ R K (20)
By choosing


10
Q= , R = 1e − 7 (21)
01
is obtained the gain vector
 
K = 9.1627e3 9.9472e3 (22)

Non-linear Control Algorithm Design

This design consists in the definition of the position error (23), the proposition of a
pole assignable second order desired dynamic (24), and finally the substitution of this
dynamic in the non-linear model yielding the control signal (25).
e = yφ − ȳφ (23)

ë(t) + k1 ė(t) + k2 e(t) + k3 ∫ e(t)dt = 0 (24)

 
1
u= −k1 ė(t) − k2 e(t) − k3 ∫ e(t)dt + a1 ẏ + a2 sen(y) + ȳ¨φ (25)
a3
The proposed dynamic (24) is stable if
k1 > 0, k2 > 0, k3 > 0 (26)
So the control gains are chosen in terms of the model parameters
k1 = 3a1 , k2 = 10a2 , k3 = 0.5a1 (27)
320 J. Díaz-Salgado et al.

4.2 Base Controllers

State Feedback
The state feedback controller proposes a control input as:
u = −K x + ki r (28)
This control applied in the system dynamics yields:
ẋ = Ax + Bu = (A − B K )x + Bkir (29)
The desired dynamics is proposed in terms of the not dominant pole p as:
(s + 25 p)(s + 250 p) (30)
The proposed dynamics is given by:
 
 s + p + k1 k2 
  = s 2 + (k1 + p)s + k2 (31)
 −1 s 

Optimal Control
The methodology design is the same of the base optimal control, the only difference is
the proposition of the R value.
R = [9e − 6] (32)

Proportional + Integral + Derivative Control

The PID transfer function commonly established in terms of proportional, integral and
derivative gains here is written in terms of the arbitrary position of two zeroes [z 1 , z 2 ]
and a global gain kd .
 
k
kd s 2 + kdp s + kkdi kd (s + z 1 )(s + z 2 )
= (33)
s s
The proposed methodology considers the design in two stages [10]: (i) the PD com-
ponent design to achieve the desired transient response (that fixes one of the zeroes) and
(ii) the PI component design to achieve the desired steady state response (that fixes the
other zero and the global gain).
For the PD component a zero z 1 is proposed one hundred times quicklier than the
not dominant pole in open loop response.
z 1 = 100 p (34)
Meanwhile for the PI component a zero z 2 is proposed to minimize the transient
response effect of the PD component.
p
z2 = p − (35)
10
The described method yields the (xx) control gains.
kd = 1e5, ki = 2e5, k p = 3e5 (36)
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 321

5 Simulation Results
Two different scenarios are considered for the simulations: nominal and perturbed con-
ditions. Nominal simulations considers the operation of the PST without perturbations
while perturbed simulations considers the NW condition with a wind velocity of 22.2 m/s.
The solar azimuth and elevation angles are calculated using a well-known sun position
algorithm [11] for Juchitán de Zaragoza, Oaxaca (16.5988373, −95.043291) on august
5 (Fig. 4).

Fig. 4. Solar azimuth and elevation angles calculated with [11] sun position algorithm.

Fig. 5. PST dish (elevation angle) transient response with proposed controllers.
322 J. Díaz-Salgado et al.

5.1 Nominal Conditions

Transient and steady state responses are simulated for base and parabolic dish pro-
posed controllers. In Table 2 the performance parameters of the transient response of the
parabolic dish (elevation angle) (Fig. 5) and base (azimuth angle) (Fig. 6) are presented.

Fig. 6. PST base (azimuth angle) transient response with proposed controllers.

The comparison shows a better performance of non-linear and PID control for dish
and base controllers respectively.

Table 2. Performance of dish and base transient responses.

 
Parameter ts [s] %O S[/] ess ◦
Dish (elevation angle) control State feedback + Integral control 15 10 0.22
Optimal control 11 13 0.23
Non-linear control 5 7 0.15
Base (azimuth angle) control State feedback 45 0 11
Optimal control 18 12 2
PID control 2 11 0

The steady state trajectory integral error of dish (Fig. 7) and base (Fig. 8) controllers
shows that non-linear and PID controllers respectively present the lower steady state
error.
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 323

Fig. 7. PST dish (elevation angle) steady state integral error response with proposed controllers.

Fig. 8. PST base (azimuth angle) steady state integral error response with proposed controllers.

5.2 Perturbed Conditions

As non-linear and PID controllers for the dish and base controllers respectively shown the
better transient and steady state performance, the perturbed conditions are only proved
with this selected controllers.
The perturbed input signals (Fig. 9) are obtained with a 6th order polynomial
approximation of the hourly moment coefficients generated in the CFD simulations.
The sun trajectory tracking along the test day (Fig. 10) shows good agreement
between desired and actual elevation and azimuth angles with the selected designed
controllers with perturbed conditions, as the steady state tracking errors (Fig. 11) stays
in an order of 10−5 in both cases.
324 J. Díaz-Salgado et al.

Fig. 9. PST moment dish and base perturbations considered for perturbed conditions.

Fig. 10. Trajectory tracking of dish and base controllers with perturbed conditions

Finally the comparison between nominal and perturbed control signals (Fig. 12)
shows a huge difference between the nominal and perturbed control signals which jus-
tifies: (i) the need to incorporate a closed loop control scheme in order to minimize the
wind load deviation, and (ii) gives an insight of the correct dimensioning of the system
actuators under perturbed NW condition.
 A Closed Loop Controller Design Which Can Minimize the Wind Effects 325

Fig. 11. Trajectory tracking errors of dish and base controllers with perturbed conditions

Fig. 12. Comparison between nominal and perturbed control signals

6 Conclusions
In this work, a (CFD) wind flow analysis technique was applied to model a North Wind
weather condition to obtain the aerodynamic momentum coefficients of a Parabolic Solar
Tracker for different azimuth and elevation angles along a test day. The proposition
of a numerical and analytical model of the PST in conjunction with the aerodynamic
momentum coefficients, obtained in the CFD analysis, permits to (i) establish a dynamic
analysis and (ii) design a closed loop controller to minimize the effect of the wind. The
closed loop responses show that the control torque in worst cases must increase two to
three times the nominal torque (without wind) to counteract the aerodynamic forces.
The results of the present work can be used to (i) correct sizing of the control system
actuators for a specific wind condition and location and (ii) design a lighter weighted
326 J. Díaz-Salgado et al.

structure because a great part of the disturbance rejection of the wind load can be done
by the actuator of the control system, instead of an increment of the mechanical structure
robustness. The presented results suggest that the nominal power required to control a
PST without aerodynamic loads is typically underestimated.

References
1. Baharoon DA, Rahman HA, Wan Omar WZ, y Fadhl SO (2015) Historical development
of concentrating solar power technologies to generate clean electricity efficiently a review.
Renew Sustain Energy Rev 41:996–1027
2. Schiel W, Keck T (2012) Parabolic dish concentrating solar power (CSP) systems, Schlaich
Bergermann und Partner, Germany, pp 284–322
3. Ruelas Ruiz JE, Cota A, Ochoa F, Lucero B, Burgos T, Delfin J, Soto A, Tzab J, Lopez
J, Nañez J (2014) Structural design methodology for solar concentrators subjected to wind
loads. Energy Procedia 57:2872–2878
4. Roschke EJ (1984) Wind loading on solar concentrators: some general considerations. Wiley,
New Jersey
5. Paetzold J, Cochard S, Fletcher D, Vassallo A (2015) Wind engineering analysis of parabolic
trough collectors to optimise wind loads and heat loss. Energy Procedia 69:168–177
6. Meteored Homepage. https://www.meteored.mx/noticias/actualidad/evento-de-norte-en-
mexico.html. Accessed 10 Jan 2020
7. Hijazi H, Ossama Mokhiamar OE (2016) Mechanical design of a low cost parabolic solar
dish concentrator. Alexandria Eng J 55:1–11
8. Reyes Cortés F (2012) Matlab Aplicado a Robótica y Mecatrónica. Alfaomega
9. Huerta Chavez O, Díaz Salgado J, Ortiz Cil C, Jimenez Escalona J, Torres Cedillo S (2019)
Cuamatzi R (2019) 3D CFD wind flow analysis technique applied to a parabolic solar tracker
for two extreme weather conditions with experimental results and a controller proposition.
Renew Sustain Energy 11:023702
10. Nise N (2011) Control systems engineering. John Wiley & Sons
11. Reda I, Andreas A (2003) Solar position algorithm for solar radiation application. National
Renewable Energy Laboratory (NREL) Technical report NREL/TP-560-34302
Author Index

A D
Abundis-Fong, H. F., 179 Dávila-Martínez, Edgar Arturo, 261
Acevedo, Mario, 25 de Jesús Trujillo Romero, Felipe, 290
Adrian, Bolaños, 218 Diaz-Martinez, Juan G., 159
Ahmad, Rafiq, 44, 53 Díaz-Ponce, A., 170
Alfredo Arias, M., 85 Díaz-Salgado, Jorge, 109, 261, 310
Alvarado Requena, Erick, 138 Diveev, Askhat, 279
Álvarez-Gallegos, J., 97 Dohnal, Fadi, 193
Amador, Jhoset Méndez, 228 Dubrovin, Grigory, 279
Angulo, M., 170
Arias-Montaño, A., 119, 147 E
Arias-Montiel, Manuel, 249 Elias, Noe Reyes, 138
Arteaga-Escamilla, C.-A., 97 Enciso-Contreras, E., 310
Estrada, Antonio, 138
B
F
Baena, Adán H., 290
Flores, Ernesto, 206
Bernal-Orozco, R. A., 147
G
C Gaponenko, Elena, 36
Castorena Solache, I., 119 García-Barrera, Jesús, 109
Castro, Gibran Jalil Garnica, 76 Garnica-Castro, G. J., 269
Castro-Linares, R., 97 Gegov, Alexander, 9, 15
Ceccarelli, Marco, 249 Genaro Carreon, C., 85
Chairez, I., 240 Gómez-Mancilla, J.-C., 179
Chavez, León Felipe Guevara, 1 Guevara-Morales, G., 119
Chávez, Oliver M. Huerta, 65, 76 Gutierrez-Carmona, Irandi, 159
Collado, Joaquín, 193
Contreras, Miguel A., 9 H
Cortés-Marín, L. A., 310 Hernández, Eusebio Eduardo, 304
Cruz, Jaime Cruz, 65, 76 Hernández-Cortés, T., 179
Cruz-Cruz, J., 269 Hernandez-Maldonado, Ana, 159
Cruz-Ortiz, D., 240 Hernández-Rodríguez, Felipe, 15

© The Editor(s) (if applicable) and The Author(s), under exclusive license
to Springer Nature Switzerland AG 2020
E. E. Hernandez et al. (Eds.): LASIRS 2019, MMS 86, pp. 327–328, 2020.
https://doi.org/10.1007/978-3-030-45402-9
328 Author Index

Herrera-Cordero, Mario E., 249 P

Huerta, C. Oliver M., 85 Poznyak, A., 240
Huerta-Chávez, O., 147 Poznyak, Alexander, 206
Huerta-Chávez, Oliver Marcel, 109, 119, 261,
269, 310 R
Ramírez-Barrios, Miguel, 193
I Ramírez, Gengis Toledo, 138
Irandi, Gutiérrez-Carmona, 218 Razvarz, Sina, 1, 15
Rivera-Arreola, D.-E., 179
Rivera-Blas, Raúl, 228
J
Rodríguez, Berenice, 138
Jafari, Raheleh, 9, 15
Rybak, Larisa, 36
Jia, Feiyu, 53
Jimenez, E. A. Villegas, 310
S
Juan, Gabino, 218
Sanchez, Bonifacio, 206
Sánchez-López, Rafael, 109
K
Sergio Torres, C., 85
Keshtkar, Sajjad, 159, 170
Kevin, Sánchez, 218
T
Kojima, Hirohisa, 159
Toledo-Velázquez, M., 269
Tzintzun, Juan, 53
L
López, Rafael Sánchez, 76 U
Lugo-González, Esther, 249 Uribe, Jorge, 138

M V
Macedo-Chagolla, Fernando, 261 Valdez, Sergio I., 290
Malyshev, Dmitry, 36, 279 Valdivia, R., 170
Manuel, Martínez, 218 Valentín, L., 170
Montes, Moisés M., 290 Vargas-Jarillo, Cristóbal, 1
Velázquez, Miguel Toledo, 65, 76
Villegas-Jiménez, Ernesto Alonso, 261
N
Virgen-Vázquez, Christopher, 304
Niño-Suárez, P. A., 228
Nozdracheva, Anna, 279
W
Wang, Tianxiong, 44
O
Ordaz, Patricio, 206 Y
Ortega H., José Ángel L., 85 Yu, Wen, 9
Ortiz-Cil, C., 310
Orvañanos-Guerrero, María T., 25 Z
Osornio, Armando Oropeza, 304 Zirion Flores, M., 119